diff --git a/src/tokenizer/index.js b/src/tokenizer/index.js
index b41067c43b..743c3286a5 100644
--- a/src/tokenizer/index.js
+++ b/src/tokenizer/index.js
@@ -17,18 +17,24 @@ import State from "./state";
 // The following character codes are forbidden from being
 // an immediate sibling of NumericLiteralSeparator _
 
-const forbiddenNumericLiteralSeparatorSiblings = [
-  46,  // .
-  66,  // B
-  69,  // E
-  79,  // O
-  88,  // X
-  95,  // _ (multiple separators are not allowed)
-  98,  // b
-  101, // e
-  111, // o
-  120, // x
-];
+const forbiddenNumericSeparatorSiblings = {
+  decBinOct: [
+    46,  // .
+    66,  // B
+    69,  // E
+    79,  // O
+    95,  // _ (multiple separators are not allowed)
+    98,  // b
+    101, // e
+    111, // o
+  ],
+  hex: [
+    46,  // .
+    88,  // X
+    95,  // _ (multiple separators are not allowed)
+    120, // x
+  ],
+};
 
 // Object type used to represent tokens. Note that normally, tokens
 // simply exist as properties on the parser object. This is only
@@ -567,6 +573,9 @@ export default class Tokenizer extends LocationParser {
 
   readInt(radix: number, len?: number): number | null {
     const start = this.state.pos;
+    const forbiddenSiblings = radix === 16 ?
+      forbiddenNumericSeparatorSiblings.hex :
+      forbiddenNumericSeparatorSiblings.decBinOct;
     let total = 0;
 
     for (let i = 0, e = len == null ? Infinity : len; i < e; ++i) {
@@ -577,8 +586,8 @@ export default class Tokenizer extends LocationParser {
         const prev = this.input.charCodeAt(this.state.pos - 1);
         const next = this.input.charCodeAt(this.state.pos + 1);
         if (code === 95) {
-          if ((forbiddenNumericLiteralSeparatorSiblings.indexOf(prev) > -1) ||
-              (forbiddenNumericLiteralSeparatorSiblings.indexOf(next) > -1) ||
+          if ((forbiddenSiblings.indexOf(prev) > -1) ||
+              (forbiddenSiblings.indexOf(next) > -1) ||
               Number.isNaN(next)) {
             this.raise(this.state.pos, "Invalid NumericLiteralSeparator");
           }
diff --git a/test/fixtures/experimental/numeric-literal-separator/valid-12/actual.js b/test/fixtures/experimental/numeric-literal-separator/valid-12/actual.js
new file mode 100644
index 0000000000..4115d0bd97
--- /dev/null
+++ b/test/fixtures/experimental/numeric-literal-separator/valid-12/actual.js
@@ -0,0 +1 @@
+0xBE_be_EB_eb;
diff --git a/test/fixtures/experimental/numeric-literal-separator/valid-12/expected.json b/test/fixtures/experimental/numeric-literal-separator/valid-12/expected.json
new file mode 100644
index 0000000000..8719155cde
--- /dev/null
+++ b/test/fixtures/experimental/numeric-literal-separator/valid-12/expected.json
@@ -0,0 +1,69 @@
+{
+  "type": "File",
+  "start": 0,
+  "end": 14,
+  "loc": {
+    "start": {
+      "line": 1,
+      "column": 0
+    },
+    "end": {
+      "line": 1,
+      "column": 14
+    }
+  },
+  "program": {
+    "type": "Program",
+    "start": 0,
+    "end": 14,
+    "loc": {
+      "start": {
+        "line": 1,
+        "column": 0
+      },
+      "end": {
+        "line": 1,
+        "column": 14
+      }
+    },
+    "sourceType": "script",
+    "body": [
+      {
+        "type": "ExpressionStatement",
+        "start": 0,
+        "end": 14,
+        "loc": {
+          "start": {
+            "line": 1,
+            "column": 0
+          },
+          "end": {
+            "line": 1,
+            "column": 14
+          }
+        },
+        "expression": {
+          "type": "NumericLiteral",
+          "start": 0,
+          "end": 13,
+          "loc": {
+            "start": {
+              "line": 1,
+              "column": 0
+            },
+            "end": {
+              "line": 1,
+              "column": 13
+            }
+          },
+          "extra": {
+            "rawValue": 3200183275,
+            "raw": "0xBE_be_EB_eb"
+          },
+          "value": 3200183275
+        }
+      }
+    ],
+    "directives": []
+  }
+}
\ No newline at end of file