diff --git a/Cargo.lock b/Cargo.lock
index 95093c1..0ef920d 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -740,6 +740,15 @@ version = "1.15.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "48c757948c5ede0e46177b7add2e67155f70e33c07fea8284df6576da70b3719"
 
+[[package]]
+name = "encoding_rs"
+version = "0.8.35"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "75030f3c4f45dafd7586dd6780965a8c7e8e285a5ecb86713e63a79c5b2766f3"
+dependencies = [
+ "cfg-if",
+]
+
 [[package]]
 name = "equivalent"
 version = "1.0.2"
@@ -1699,6 +1708,7 @@ name = "obscura-net"
 version = "0.1.0"
 dependencies = [
  "async-trait",
+ "encoding_rs",
  "reqwest",
  "serde",
  "serde_json",
@@ -2165,7 +2175,7 @@ dependencies = [
  "errno",
  "libc",
  "linux-raw-sys 0.12.1",
- "windows-sys 0.52.0",
+ "windows-sys 0.61.2",
 ]
 
 [[package]]
@@ -2564,10 +2574,10 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "32497e9a4c7b38532efcdebeef879707aa9f794296a4f0244f6f69e9bc8574bd"
 dependencies = [
  "fastrand",
- "getrandom 0.3.4",
+ "getrandom 0.4.2",
  "once_cell",
  "rustix 1.1.4",
- "windows-sys 0.52.0",
+ "windows-sys 0.61.2",
 ]
 
 [[package]]
diff --git a/crates/obscura-browser/src/page.rs b/crates/obscura-browser/src/page.rs
index 0a8067a..a3e2b43 100644
--- a/crates/obscura-browser/src/page.rs
+++ b/crates/obscura-browser/src/page.rs
@@ -402,7 +402,9 @@ impl Page {
         let mut fetched: std::collections::HashMap<usize, (String, String, obscura_net::Response)> = std::collections::HashMap::new();
         for result in fetch_results {
             if let Some((idx, url, resp)) = result {
-                let code = String::from_utf8_lossy(&resp.body).to_string();
+                // Script bodies: only the HTTP Content-Type charset matters
+                // (no in-band meta-charset for JS).
+                let code = obscura_net::decode_non_html(&resp.body, resp.content_type());
                 fetched.insert(idx, (url, code, resp));
             }
         }
@@ -650,7 +652,11 @@ impl Page {
             self.url = Some(response.url.clone());
         }
 
-        let body_text = String::from_utf8_lossy(&response.body).to_string();
+        // Honor the response charset: HTTP Content-Type → <meta charset> sniff
+        // in the first 1KB → UTF-8 fallback. Without this, every non-UTF-8
+        // page (GBK, Big5, Shift-JIS, Windows-125x, EUC-KR, ISO-8859-x)
+        // came through as replacement characters.
+        let body_text = obscura_net::decode_response(&response.body, response.content_type());
         let dom = parse_html(&body_text);
 
         self.title = dom
@@ -718,7 +724,9 @@ impl Page {
         let mut css_sources = Vec::new();
         for result in css_results {
             if let Some((url_str, resp)) = result {
-                let css = String::from_utf8_lossy(&resp.body).to_string();
+                // CSS bodies: honor the Content-Type charset; CSS @charset is
+                // out of scope for the current scrape-focused pipeline.
+                let css = obscura_net::decode_non_html(&resp.body, resp.content_type());
                 self.record_network_event(&url_str, "GET", "Stylesheet", resp.status, &resp.headers, resp.body.len());
                 css_sources.push(css);
             }
diff --git a/crates/obscura-net/Cargo.toml b/crates/obscura-net/Cargo.toml
index c1889fa..328bff8 100644
--- a/crates/obscura-net/Cargo.toml
+++ b/crates/obscura-net/Cargo.toml
@@ -16,6 +16,7 @@ serde_json = { workspace = true }
 tracing = { workspace = true }
 thiserror = { workspace = true }
 async-trait = "0.1"
+encoding_rs = "0.8"
 wreq-util = { version = "3.0.0-rc.10", optional = true }
 
 # `prefix-symbols` renames BoringSSL exports to avoid clashes with system OpenSSL,
diff --git a/crates/obscura-net/src/client.rs b/crates/obscura-net/src/client.rs
index 40004fb..c971858 100644
--- a/crates/obscura-net/src/client.rs
+++ b/crates/obscura-net/src/client.rs
@@ -21,8 +21,17 @@ pub struct Response {
 }
 
 impl Response {
-    pub fn text(&self) -> Result<String, std::string::FromUtf8Error> {
-        String::from_utf8(self.body.clone())
+    /// Decode the body as text, honoring the response charset.
+    ///
+    /// Uses the HTTP `Content-Type` header's `charset=` parameter, then for
+    /// HTML responses falls back to sniffing `<meta charset>` in the first
+    /// 1KB, then UTF-8. Mirrors browser behaviour per the HTML5 spec.
+    pub fn text(&self) -> String {
+        if self.is_html() {
+            crate::encoding::decode_response(&self.body, self.content_type())
+        } else {
+            crate::encoding::decode_non_html(&self.body, self.content_type())
+        }
     }
 
     pub fn header(&self, name: &str) -> Option<&str> {
diff --git a/crates/obscura-net/src/encoding.rs b/crates/obscura-net/src/encoding.rs
new file mode 100644
index 0000000..dffe2ce
--- /dev/null
+++ b/crates/obscura-net/src/encoding.rs
@@ -0,0 +1,197 @@
+//! Charset detection and decoding for HTTP response bodies.
+//!
+//! Issue #113: obscura used to call `String::from_utf8_lossy` on every
+//! response body, which silently corrupts every non-UTF-8 page (GBK, Big5,
+//! Shift-JIS, Windows-125x, EUC-KR, ISO-8859-x). Picking the right decoder
+//! is required for scraping non-Latin sites at all.
+//!
+//! Detection order, mirroring real browsers (HTML5 spec § 8.2.2.4):
+//!   1. `Content-Type: text/html; charset=...` from the HTTP response header.
+//!   2. `<meta charset="...">` or `<meta http-equiv="Content-Type" content="text/html; charset=...">`
+//!      sniffed from the first 1024 bytes of the body.
+//!   3. Default UTF-8.
+//!
+//! For non-HTML resources (JS, CSS, JSON), only steps 1 and 3 apply.
+
+use encoding_rs::{Encoding, UTF_8};
+
+/// Decode an HTTP response body. `content_type_header` is the raw header
+/// value if present (e.g. `text/html; charset=gbk`). For HTML resources,
+/// the parser also sniffs `<meta charset>` in the first 1KB.
+pub fn decode_response(bytes: &[u8], content_type_header: Option<&str>) -> String {
+    let (encoding, _) = detect_encoding(bytes, content_type_header);
+    let (cow, _, _) = encoding.decode(bytes);
+    cow.into_owned()
+}
+
+/// Same as `decode_response` but skips the `<meta charset>` sniff. Use for
+/// non-HTML resources where embedded HTML meta tags are not authoritative
+/// (script and style bodies, JSON, plain text).
+pub fn decode_non_html(bytes: &[u8], content_type_header: Option<&str>) -> String {
+    let encoding = content_type_header
+        .and_then(charset_from_content_type)
+        .and_then(|name| Encoding::for_label(name.as_bytes()))
+        .unwrap_or(UTF_8);
+    let (cow, _, _) = encoding.decode(bytes);
+    cow.into_owned()
+}
+
+/// Resolve the encoding to use for an HTML response, mirroring the HTML5
+/// detection order. Returns the encoding and a tag describing where it was
+/// picked from (for logging / tests).
+pub fn detect_encoding<'a>(
+    bytes: &'a [u8],
+    content_type_header: Option<&str>,
+) -> (&'static Encoding, &'static str) {
+    if let Some(charset) = content_type_header.and_then(charset_from_content_type) {
+        if let Some(enc) = Encoding::for_label(charset.as_bytes()) {
+            return (enc, "content-type");
+        }
+    }
+    if let Some(enc) = sniff_meta_charset(bytes) {
+        return (enc, "meta-charset");
+    }
+    (UTF_8, "default-utf8")
+}
+
+/// Pull the `charset=` parameter out of a Content-Type header value.
+fn charset_from_content_type(header: &str) -> Option<String> {
+    for part in header.split(';') {
+        let trimmed = part.trim();
+        if let Some(rest) = trimmed.strip_prefix("charset=").or_else(|| trimmed.strip_prefix("Charset=")) {
+            // Strip surrounding quotes if present.
+            let value = rest.trim_matches(|c: char| c == '"' || c == '\'').trim();
+            if !value.is_empty() {
+                return Some(value.to_ascii_lowercase());
+            }
+        }
+        // Some servers send `Content-Type: text/html; CHARSET = gbk`.
+        let lower = trimmed.to_ascii_lowercase();
+        if let Some(rest) = lower.strip_prefix("charset") {
+            let rest = rest.trim_start();
+            if let Some(rest) = rest.strip_prefix('=') {
+                let value = rest.trim().trim_matches(|c: char| c == '"' || c == '\'');
+                if !value.is_empty() {
+                    return Some(value.to_ascii_lowercase());
+                }
+            }
+        }
+    }
+    None
+}
+
+/// Scan the first 1024 bytes for a `<meta charset="...">` or
+/// `<meta http-equiv="Content-Type" content="...; charset=...">` declaration.
+/// We only look at ASCII bytes; valid meta-charset declarations are always
+/// ASCII regardless of the page's actual encoding.
+fn sniff_meta_charset(bytes: &[u8]) -> Option<&'static Encoding> {
+    let prefix_len = bytes.len().min(1024);
+    let prefix = &bytes[..prefix_len];
+    // Lossy is fine: any meta charset attribute is ASCII, even on a non-UTF-8 page.
+    let s = String::from_utf8_lossy(prefix).to_ascii_lowercase();
+
+    // Look for any `<meta ... charset=...>` pattern in the first 1KB. We
+    // intentionally accept both the modern shorthand (`<meta charset=gbk>`)
+    // and the legacy http-equiv form (`<meta http-equiv="content-type" content="text/html; charset=gbk">`).
+    let mut pos = 0;
+    while let Some(meta_start) = s[pos..].find("<meta") {
+        let abs = pos + meta_start;
+        // Find the closing `>` for this meta tag.
+        let end = s[abs..].find('>').map(|e| abs + e).unwrap_or(s.len());
+        let tag = &s[abs..end];
+
+        if let Some(charset_pos) = tag.find("charset") {
+            let after = &tag[charset_pos + "charset".len()..];
+            let after = after.trim_start();
+            if let Some(eq_rest) = after.strip_prefix('=') {
+                let value = eq_rest
+                    .trim_start()
+                    .trim_start_matches(|c: char| c == '"' || c == '\'')
+                    .split(|c: char| c == '"' || c == '\'' || c == ';' || c.is_whitespace() || c == '/')
+                    .next()
+                    .unwrap_or("");
+                if !value.is_empty() {
+                    if let Some(enc) = Encoding::for_label(value.as_bytes()) {
+                        return Some(enc);
+                    }
+                }
+            }
+        }
+
+        pos = end + 1;
+        if pos >= s.len() {
+            break;
+        }
+    }
+
+    None
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn content_type_charset_wins() {
+        let bytes = b"<html><head><meta charset=\"utf-8\"></head><body></body></html>";
+        let (enc, source) = detect_encoding(bytes, Some("text/html; charset=gbk"));
+        assert_eq!(enc.name(), "GBK");
+        assert_eq!(source, "content-type");
+    }
+
+    #[test]
+    fn content_type_quoted_charset_is_parsed() {
+        let (enc, _) = detect_encoding(b"", Some("text/html; charset=\"Shift_JIS\""));
+        assert_eq!(enc.name(), "Shift_JIS");
+    }
+
+    #[test]
+    fn meta_charset_used_when_header_missing() {
+        let bytes = b"<!doctype html><html><head><meta charset=\"big5\"></head></html>";
+        let (enc, source) = detect_encoding(bytes, None);
+        assert_eq!(enc.name(), "Big5");
+        assert_eq!(source, "meta-charset");
+    }
+
+    #[test]
+    fn meta_http_equiv_charset_is_recognized() {
+        let bytes = b"<html><head><meta http-equiv=\"Content-Type\" content=\"text/html; charset=EUC-KR\"></head></html>";
+        let (enc, _) = detect_encoding(bytes, None);
+        assert_eq!(enc.name(), "EUC-KR");
+    }
+
+    #[test]
+    fn no_charset_anywhere_falls_back_to_utf8() {
+        let bytes = b"<html><body>hello</body></html>";
+        let (enc, source) = detect_encoding(bytes, None);
+        assert_eq!(enc.name(), "UTF-8");
+        assert_eq!(source, "default-utf8");
+    }
+
+    #[test]
+    fn decode_response_gbk_bytes_roundtrip() {
+        // "你好" (ni hao) encoded as GBK = C4 E3 BA C3
+        let bytes: &[u8] = &[0xC4, 0xE3, 0xBA, 0xC3];
+        let s = decode_response(bytes, Some("text/html; charset=gbk"));
+        assert_eq!(s, "你好");
+    }
+
+    #[test]
+    fn decode_non_html_skips_meta_sniff() {
+        // A JS body that happens to contain a string `<meta charset="gbk">`
+        // must NOT be decoded as GBK — non-HTML resources only honor the
+        // HTTP header.
+        let bytes = br#"var x = '<meta charset="gbk">'; // not the real charset"#;
+        let s = decode_non_html(bytes, Some("application/javascript"));
+        assert!(s.contains("<meta charset="));
+    }
+
+    #[test]
+    fn meta_sniff_only_scans_first_1kb() {
+        let mut bytes = vec![b' '; 2048];
+        bytes.extend_from_slice(b"<meta charset=\"gbk\">");
+        let (enc, _) = detect_encoding(&bytes, None);
+        // Beyond 1KB: ignored, fall back to UTF-8.
+        assert_eq!(enc.name(), "UTF-8");
+    }
+}
diff --git a/crates/obscura-net/src/lib.rs b/crates/obscura-net/src/lib.rs
index 6bea89b..b319d08 100644
--- a/crates/obscura-net/src/lib.rs
+++ b/crates/obscura-net/src/lib.rs
@@ -1,5 +1,6 @@
 pub mod client;
 pub mod cookies;
+pub mod encoding;
 pub mod interceptor;
 pub mod robots;
 pub mod blocklist;
@@ -8,6 +9,7 @@ pub mod wreq_client;
 
 pub use client::{ObscuraHttpClient, ObscuraNetError, RequestInfo, ResourceType, Response};
 pub use cookies::{CookieInfo, CookieJar};
+pub use encoding::{decode_non_html, decode_response};
 pub use robots::RobotsCache;
 pub use blocklist::is_blocked as is_tracker_blocked;
 #[cfg(feature = "stealth")]