diff --git a/doc/source/whatsnew/v0.23.4.txt b/doc/source/whatsnew/v0.23.4.txt
index a88c22e3d01f7..9c2ef7ecb601c 100644
--- a/doc/source/whatsnew/v0.23.4.txt
+++ b/doc/source/whatsnew/v0.23.4.txt
@@ -31,7 +31,7 @@ Bug Fixes
 
 **Conversion**
 
--
+- Bug where unwanted casting of float to int in :func:`isin` led to incorrect comparison outcome (:issue:`21804`)
 -
 
 **Indexing**
diff --git a/pandas/core/algorithms.py b/pandas/core/algorithms.py
index 6e49e8044ff25..3833917f2e419 100644
--- a/pandas/core/algorithms.py
+++ b/pandas/core/algorithms.py
@@ -23,7 +23,7 @@
     is_period_dtype,
     is_numeric_dtype, is_float_dtype,
     is_bool_dtype, needs_i8_conversion,
-    is_datetimetz,
+    is_datetimetz, is_datetime_or_timedelta_dtype,
     is_datetime64_any_dtype, is_datetime64tz_dtype,
     is_timedelta64_dtype, is_datetimelike,
     is_interval_dtype, is_scalar, is_list_like,
@@ -39,6 +39,8 @@
 from pandas.util._decorators import (Appender, Substitution,
                                      deprecate_kwarg)
 
+from pandas._libs.tslibs.timestamps import Timestamp
+
 _shared_docs = {}
 
 
@@ -415,33 +417,40 @@ def isin(comps, values):
     comps = com._values_from_object(comps)
 
     comps, dtype, _ = _ensure_data(comps)
-    values, _, _ = _ensure_data(values, dtype=dtype)
+
+    is_time_like = lambda x: (is_datetime_or_timedelta_dtype(x)
+                              or isinstance(x, Timestamp))
+
+    is_int = lambda x: ((x == np.int64) or (x == int))
+    is_float = lambda x: ((x == np.float64) or (x == float))
+
+    if is_time_like(dtype):
+        values, _, _ = _ensure_data(values, dtype=dtype)
+    else:
+        values, _, _ = _ensure_data(values)
+
+    comps_types = set(type(v) for v in comps)
+    values_types = set(type(v) for v in values)
 
     # faster for larger cases to use np.in1d
-    f = lambda x, y: htable.ismember_object(x, values)
+    f = lambda x, y: htable.ismember_object(x.astype(object), y.astype(object))
 
     # GH16012
     # Ensure np.in1d doesn't get object types or it *may* throw an exception
     if len(comps) > 1000000 and not is_object_dtype(comps):
         f = lambda x, y: np.in1d(x, y)
-    elif is_integer_dtype(comps):
-        try:
+    elif len(comps_types) == len(values_types) == 1:
+        comps_types = comps_types.pop()
+        values_types = values_types.pop()
+        if (is_int(comps_types) and is_int(values_types)):
             values = values.astype('int64', copy=False)
             comps = comps.astype('int64', copy=False)
             f = lambda x, y: htable.ismember_int64(x, y)
-        except (TypeError, ValueError):
-            values = values.astype(object)
-            comps = comps.astype(object)
-
-    elif is_float_dtype(comps):
-        try:
+        elif (is_float(comps_types) and is_float(values_types)):
             values = values.astype('float64', copy=False)
             comps = comps.astype('float64', copy=False)
             checknull = isna(values).any()
             f = lambda x, y: htable.ismember_float64(x, y, checknull)
-        except (TypeError, ValueError):
-            values = values.astype(object)
-            comps = comps.astype(object)
 
     return f(comps, values)
 
diff --git a/pandas/tests/test_algos.py b/pandas/tests/test_algos.py
index 25e64aa82cc36..8ab907a9723bd 100644
--- a/pandas/tests/test_algos.py
+++ b/pandas/tests/test_algos.py
@@ -509,42 +509,23 @@ def test_invalid(self):
         pytest.raises(TypeError, lambda: algos.isin(1, [1]))
         pytest.raises(TypeError, lambda: algos.isin([1], 1))
 
-    def test_basic(self):
-
-        result = algos.isin([1, 2], [1])
-        expected = np.array([True, False])
-        tm.assert_numpy_array_equal(result, expected)
-
-        result = algos.isin(np.array([1, 2]), [1])
-        expected = np.array([True, False])
-        tm.assert_numpy_array_equal(result, expected)
-
-        result = algos.isin(Series([1, 2]), [1])
-        expected = np.array([True, False])
-        tm.assert_numpy_array_equal(result, expected)
-
-        result = algos.isin(Series([1, 2]), Series([1]))
-        expected = np.array([True, False])
-        tm.assert_numpy_array_equal(result, expected)
-
-        result = algos.isin(Series([1, 2]), set([1]))
-        expected = np.array([True, False])
-        tm.assert_numpy_array_equal(result, expected)
-
-        result = algos.isin(['a', 'b'], ['a'])
-        expected = np.array([True, False])
-        tm.assert_numpy_array_equal(result, expected)
-
-        result = algos.isin(Series(['a', 'b']), Series(['a']))
-        expected = np.array([True, False])
-        tm.assert_numpy_array_equal(result, expected)
-
-        result = algos.isin(Series(['a', 'b']), set(['a']))
-        expected = np.array([True, False])
-        tm.assert_numpy_array_equal(result, expected)
-
-        result = algos.isin(['a', 'b'], [1])
-        expected = np.array([False, False])
+    @pytest.mark.parametrize("comps,values,expected", [
+        ([1, 2], [1], [True, False]),
+        ([1, 0], [1, 0.5], [True, False]),
+        ([1.0, 0], [1, 0.5], [True, False]),
+        ([1.0, 0.0], [1, 0], [True, True]),
+        (np.array([1, 2]), [1], [True, False]),
+        (Series([1, 2]), [1], [True, False]),
+        (Series([1, 2]), Series([1]), [True, False]),
+        (Series([1, 2]), set([1]), [True, False]),
+        (['a', 'b'], ['a'], [True, False]),
+        (Series(['a', 'b']), Series(['a']), [True, False]),
+        (Series(['a', 'b']), set(['a']), [True, False]),
+        (['a', 'b'], [1], [False, False])
+    ])
+    def test_basic(self, comps, values, expected):
+        result = algos.isin(comps, values)
+        expected = np.array(expected)
         tm.assert_numpy_array_equal(result, expected)
 
     def test_i8(self):