diff --git a/asv_bench/benchmarks/timeseries.py b/asv_bench/benchmarks/timeseries.py
index efe713639fec9..b7151ad2eaa99 100644
--- a/asv_bench/benchmarks/timeseries.py
+++ b/asv_bench/benchmarks/timeseries.py
@@ -510,3 +510,17 @@ def time_begin_incr_rng(self):
 
     def time_begin_decr_rng(self):
         self.rng - self.semi_month_begin
+
+
+class DatetimeAccessor(object):
+    def setup(self):
+        self.N = 100000
+        self.series = pd.Series(
+            pd.date_range(start='1/1/2000', periods=self.N, freq='T')
+        )
+
+    def time_dt_accessor(self):
+        self.series.dt
+
+    def time_dt_accessor_normalize(self):
+        self.series.dt.normalize()
diff --git a/doc/source/whatsnew/v0.21.0.txt b/doc/source/whatsnew/v0.21.0.txt
index 8b2c4d16f4e1a..e21ee8d7d31f5 100644
--- a/doc/source/whatsnew/v0.21.0.txt
+++ b/doc/source/whatsnew/v0.21.0.txt
@@ -296,6 +296,7 @@ Performance Improvements
 ~~~~~~~~~~~~~~~~~~~~~~~~
 
 - Improved performance of instantiating :class:`SparseDataFrame` (:issue:`16773`)
+- :attr:`Series.dt` no longer performs frequency inference, yielding a large speedup when accessing the attribute (:issue:`17210`)
 
 
 .. _whatsnew_0210.bug_fixes:
diff --git a/pandas/core/indexes/accessors.py b/pandas/core/indexes/accessors.py
index ce3143b342cec..88297ac70984d 100644
--- a/pandas/core/indexes/accessors.py
+++ b/pandas/core/indexes/accessors.py
@@ -61,23 +61,20 @@ def maybe_to_datetimelike(data, copy=False):
         data = orig.values.categories
 
     if is_datetime64_dtype(data.dtype):
-        return DatetimeProperties(DatetimeIndex(data, copy=copy, freq='infer'),
+        return DatetimeProperties(DatetimeIndex(data, copy=copy),
                                   index, name=name, orig=orig)
     elif is_datetime64tz_dtype(data.dtype):
-        return DatetimeProperties(DatetimeIndex(data, copy=copy, freq='infer',
-                                                ambiguous='infer'),
+        return DatetimeProperties(DatetimeIndex(data, copy=copy),
                                   index, data.name, orig=orig)
     elif is_timedelta64_dtype(data.dtype):
-        return TimedeltaProperties(TimedeltaIndex(data, copy=copy,
-                                                  freq='infer'), index,
+        return TimedeltaProperties(TimedeltaIndex(data, copy=copy), index,
                                    name=name, orig=orig)
     else:
         if is_period_arraylike(data):
             return PeriodProperties(PeriodIndex(data, copy=copy), index,
                                     name=name, orig=orig)
         if is_datetime_arraylike(data):
-            return DatetimeProperties(DatetimeIndex(data, copy=copy,
-                                                    freq='infer'), index,
+            return DatetimeProperties(DatetimeIndex(data, copy=copy), index,
                                       name=name, orig=orig)
 
     raise TypeError("cannot convert an object of type {0} to a "
@@ -162,6 +159,10 @@ class DatetimeProperties(Properties):
     def to_pydatetime(self):
         return self.values.to_pydatetime()
 
+    @property
+    def freq(self):
+        return self.values.inferred_freq
+
 
 DatetimeProperties._add_delegate_accessors(
     delegate=DatetimeIndex,
@@ -202,6 +203,10 @@ def components(self):
         """
         return self.values.components.set_index(self.index)
 
+    @property
+    def freq(self):
+        return self.values.inferred_freq
+
 
 TimedeltaProperties._add_delegate_accessors(
     delegate=TimedeltaIndex,