diff --git a/packages/grpc-js/package.json b/packages/grpc-js/package.json
index 1dd0f0cbe..b11495c6f 100644
--- a/packages/grpc-js/package.json
+++ b/packages/grpc-js/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@grpc/grpc-js",
-  "version": "1.12.4",
+  "version": "1.12.5",
   "description": "gRPC Library for Node - pure JS implementation",
   "homepage": "https://grpc.io/",
   "repository": "https://github.com/grpc/grpc-node/tree/master/packages/grpc-js",
diff --git a/packages/grpc-js/src/subchannel-call.ts b/packages/grpc-js/src/subchannel-call.ts
index 6fc1e1eb5..d2b5f076a 100644
--- a/packages/grpc-js/src/subchannel-call.ts
+++ b/packages/grpc-js/src/subchannel-call.ts
@@ -190,7 +190,22 @@ export class Http2SubchannelCall implements SubchannelCall {
       try {
         messages = this.decoder.write(data);
       } catch (e) {
-        this.cancelWithStatus(Status.RESOURCE_EXHAUSTED, (e as Error).message);
+        /* Some servers send HTML error pages along with HTTP status codes.
+         * When the client attempts to parse this as a length-delimited
+         * message, the parsed message size is greater than the default limit,
+         * resulting in a message decoding error. In that situation, the HTTP
+         * error code information is more useful to the user than the
+         * RESOURCE_EXHAUSTED error is, so we report that instead. Normally,
+         * we delay processing the HTTP status until after the stream ends, to
+         * prioritize reporting the gRPC status from trailers if it is present,
+         * but when there is a message parsing error we end the stream early
+         * before processing trailers. */
+        if (this.httpStatusCode !== undefined && this.httpStatusCode !== 200) {
+          const mappedStatus = mapHttpStatusCode(this.httpStatusCode);
+          this.cancelWithStatus(mappedStatus.code, mappedStatus.details);
+        } else {
+          this.cancelWithStatus(Status.RESOURCE_EXHAUSTED, (e as Error).message);
+        }
         return;
       }