diff --git a/e2e/benchmarks/local-benchmark/index.js b/e2e/benchmarks/local-benchmark/index.js
index 5e90662d27..4328b129fa 100644
--- a/e2e/benchmarks/local-benchmark/index.js
+++ b/e2e/benchmarks/local-benchmark/index.js
@@ -21,6 +21,7 @@ const BACKEND_FLAGS_MAP = {
   wasm: [
     'WASM_HAS_SIMD_SUPPORT',
     'WASM_HAS_MULTITHREAD_SUPPORT',
+    'WASM_THREAD_POOL_SIZE',
     'CHECK_COMPUTATION_FOR_ERRORS',
   ],
   webgl: [
diff --git a/tfjs-backend-wasm/src/backend_wasm.ts b/tfjs-backend-wasm/src/backend_wasm.ts
index b229ad83f4..fafe486f3c 100644
--- a/tfjs-backend-wasm/src/backend_wasm.ts
+++ b/tfjs-backend-wasm/src/backend_wasm.ts
@@ -45,6 +45,12 @@ export class BackendWasm extends KernelBackend {
   constructor(public wasm: BackendWasmModule | BackendWasmThreadedSimdModule) {
     super();
     this.wasm.tfjs.init();
+    // Register the used thread pool size flag. Done it here to avoid circular
+    // import.
+    env().registerFlag('WASM_THREAD_POOL_SIZE', () => {
+      return this.getThreadPoolSize();
+    });
+
     this.dataIdMap = new DataStorage(this, engine());
   }
 
@@ -158,6 +164,11 @@ export class BackendWasm extends KernelBackend {
     return this.dataIdMap.get(dataId).memoryOffset;
   }
 
+  // Return the used thread pool size.
+  getThreadPoolSize(): number {
+    return this.wasm.tfjs.getThreadPoolSize();
+  }
+
   dispose() {
     this.wasm.tfjs.dispose();
     if ('PThread' in this.wasm) {
@@ -346,6 +357,7 @@ export async function init(): Promise<{wasm: BackendWasmModule}> {
       // Using the tfjs namespace to avoid conflict with emscripten's API.
       module.tfjs = {
         init: module.cwrap('init', null, []),
+        getThreadPoolSize: module.cwrap('get_thread_pool_size', 'number', []),
         registerTensor: module.cwrap(
             'register_tensor', null,
             [
diff --git a/tfjs-backend-wasm/src/cc/backend.cc b/tfjs-backend-wasm/src/cc/backend.cc
index 72f2428c7f..a3cac7c24c 100644
--- a/tfjs-backend-wasm/src/cc/backend.cc
+++ b/tfjs-backend-wasm/src/cc/backend.cc
@@ -62,8 +62,9 @@ int num_cores = 1;
 
 int min_num_threads = 1;
 int max_num_threads = 4;
-pthreadpool *threadpool = pthreadpool_create(
-    std::min(std::max(num_cores, min_num_threads), max_num_threads));
+int thread_pool_size =
+    std::min(std::max(num_cores, min_num_threads), max_num_threads);
+pthreadpool *threadpool = pthreadpool_create(thread_pool_size);
 
 // Registers a disposal callback for a tensor id with a given callback function.
 void register_disposal_callback(const size_t tensor_id,
@@ -90,6 +91,11 @@ EMSCRIPTEN_KEEPALIVE
 #endif
 void init() { xnn_initialize(nullptr); }
 
+#ifdef __EMSCRIPTEN__
+EMSCRIPTEN_KEEPALIVE
+#endif
+const size_t get_thread_pool_size() { return backend::thread_pool_size; }
+
 #ifdef __EMSCRIPTEN__
 EMSCRIPTEN_KEEPALIVE
 #endif
diff --git a/tfjs-backend-wasm/src/cc/backend.h b/tfjs-backend-wasm/src/cc/backend.h
index e67d3b0601..70af87ccd2 100644
--- a/tfjs-backend-wasm/src/cc/backend.h
+++ b/tfjs-backend-wasm/src/cc/backend.h
@@ -103,6 +103,9 @@ void init();
 void register_tensor(const size_t tensor_id, const size_t size,
                      void *memory_offset);
 
+// Return the used thread pool size.
+const size_t get_thread_pool_size();
+
 // Disposes the internal bookeeping for a given tensor ID.
 void dispose_data(const size_t tensor_id);
 
diff --git a/tfjs-backend-wasm/src/index_test.ts b/tfjs-backend-wasm/src/index_test.ts
index ce0ef4f612..433abbcafe 100644
--- a/tfjs-backend-wasm/src/index_test.ts
+++ b/tfjs-backend-wasm/src/index_test.ts
@@ -28,6 +28,13 @@ import {BackendWasm, setWasmPath, setWasmPaths} from './index';
  * 'wasm' so that they are always included in the test runner. See
  * `env.specFilter` in `setup_test.ts` for details.
  */
+describeWithFlags('wasm thread pool', ALL_ENVS, () => {
+  it('thread pool size', async () => {
+    const threadPoolSize = tf.env().getNumber('WASM_THREAD_POOL_SIZE');
+    expect(threadPoolSize).toBeGreaterThanOrEqual(1);
+  });
+});
+
 describeWithFlags('wasm read/write', ALL_ENVS, () => {
   it('write and read values', async () => {
     const x = tf.tensor1d([1, 2, 3]);
diff --git a/tfjs-backend-wasm/wasm-out/tfjs-backend-wasm.d.ts b/tfjs-backend-wasm/wasm-out/tfjs-backend-wasm.d.ts
index 2af757f8bc..bbcbf36c49 100644
--- a/tfjs-backend-wasm/wasm-out/tfjs-backend-wasm.d.ts
+++ b/tfjs-backend-wasm/wasm-out/tfjs-backend-wasm.d.ts
@@ -19,6 +19,8 @@ export interface BackendWasmModule extends EmscriptenModule {
   // Using the tfjs namespace to avoid conflict with emscripten's API.
   tfjs: {
     init(): void,
+    // Return the used thread pool size.
+    getThreadPoolSize(): number,
     registerTensor(id: number, size: number, memoryOffset: number): void,
     // Disposes the data behind the data bucket.
     disposeData(id: number): void,