diff --git a/datadog/dogstatsd/base.py b/datadog/dogstatsd/base.py
index 1bdb47f5d..aac71d85d 100644
--- a/datadog/dogstatsd/base.py
+++ b/datadog/dogstatsd/base.py
@@ -27,7 +27,7 @@
 from datadog.dogstatsd.route import get_default_route
 from datadog.dogstatsd.container import ContainerID
 from datadog.util.compat import is_p3k, text
-from datadog.util.format import normalize_tags
+from datadog.util.format import normalize_tags, normalize_metric_name
 from datadog.version import __version__
 
 # Logging
@@ -318,7 +318,7 @@ def __init__(
             constant_tags = []
         self.constant_tags = constant_tags + env_tags
         if namespace is not None:
-            namespace = text(namespace)
+            namespace = normalize_metric_name(text(namespace))
         self.namespace = namespace
         self.use_ms = use_ms
         self.default_sample_rate = default_sample_rate
@@ -763,7 +763,7 @@ def _serialize_metric(self, metric, metric_type, value, tags, sample_rate=1):
         # Create/format the metric packet
         return "%s%s:%s|%s%s%s%s" % (
             (self.namespace + ".") if self.namespace else "",
-            metric,
+            normalize_metric_name(metric),
             value,
             metric_type,
             ("|@" + text(sample_rate)) if sample_rate != 1 else "",
diff --git a/datadog/util/format.py b/datadog/util/format.py
index f6b1e96af..554ec2d71 100644
--- a/datadog/util/format.py
+++ b/datadog/util/format.py
@@ -10,7 +10,8 @@
 from datadog.util.compat import conditional_lru_cache
 
 TAG_INVALID_CHARS_RE = re.compile(r"[^\w\d_\-:/\.]", re.UNICODE)
-TAG_INVALID_CHARS_SUBS = "_"
+METRIC_NAME_INVALID_CHARS_RE = re.compile(r"[^\w\d_\.]", re.UNICODE)
+INVALID_CHARS_SUBS = "_"
 
 
 def pretty_json(obj):
@@ -33,10 +34,15 @@ def force_to_epoch_seconds(epoch_sec_or_dt):
 
 @conditional_lru_cache
 def _normalize_tags_with_cache(tag_list):
-    return [TAG_INVALID_CHARS_RE.sub(TAG_INVALID_CHARS_SUBS, tag) for tag in tag_list]
+    return [TAG_INVALID_CHARS_RE.sub(INVALID_CHARS_SUBS, tag) for tag in tag_list]
 
 
 def normalize_tags(tag_list):
     # We have to turn our input tag list into a non-mutable tuple for it to
     # be hashable (and thus usable) by the @lru_cache decorator.
     return _normalize_tags_with_cache(tuple(tag_list))
+
+
+@conditional_lru_cache
+def normalize_metric_name(metric_name):
+    return METRIC_NAME_INVALID_CHARS_RE.sub(INVALID_CHARS_SUBS, metric_name)
diff --git a/tests/unit/dogstatsd/test_statsd.py b/tests/unit/dogstatsd/test_statsd.py
index 658804a19..48c4e32a1 100644
--- a/tests/unit/dogstatsd/test_statsd.py
+++ b/tests/unit/dogstatsd/test_statsd.py
@@ -294,6 +294,10 @@ def test_gauge(self):
         self.statsd.gauge('gauge', 123.4)
         self.assert_equal_telemetry('gauge:123.4|g\n', self.recv(2))
 
+    def test_gauge_with_unescaped_name(self):
+        self.statsd.gauge('my|new.p3rfect#_gauge', 123.4)
+        self.assert_equal_telemetry('my_new.p3rfect__gauge:123.4|g\n', self.recv(2))
+
     def test_counter(self):
         self.statsd.increment('page.views')
         self.statsd.flush()
diff --git a/tests/unit/util/test_format.py b/tests/unit/util/test_format.py
index dc4d6b62b..9ab5d9a1d 100644
--- a/tests/unit/util/test_format.py
+++ b/tests/unit/util/test_format.py
@@ -6,7 +6,7 @@
 
 import pytest
 
-from datadog.util.format import construct_url, normalize_tags
+from datadog.util.format import construct_url, normalize_tags, normalize_metric_name
 
 
 class TestConstructURL:
@@ -52,3 +52,19 @@ class TestNormalizeTags:
     @pytest.mark.parametrize("original_tags,expected_tags", test_data)
     def test_normalize_tags(self, original_tags, expected_tags):
             assert normalize_tags(original_tags) == expected_tags
+
+class TestNormalizeMetricName:
+    """
+    Test of the format's `normalize_metric_name` functionality
+    """
+    test_data = [
+        ('', ''),
+        ('just a metric name', 'just_a_metric_name'),
+        ('xyz.abc!@#$%^&*()0987654321{}}{', 'xyz.abc__________0987654321____'),
+        ('xyz.abc_123', 'xyz.abc_123'),
+        ('абśжż西アطر', 'абśжż西アطر'),
+        ('a😃😃b', 'a__b'),
+    ]
+    @pytest.mark.parametrize("original_metric_name,expected_metric_name", test_data)
+    def test_normalize_metric_name(self, original_metric_name, expected_metric_name):
+            assert normalize_metric_name(original_metric_name) == expected_metric_name