diff --git a/doc/source/whatsnew/v0.24.0.txt b/doc/source/whatsnew/v0.24.0.txt
index d2d5d40393b62..64dc9ea51eadd 100644
--- a/doc/source/whatsnew/v0.24.0.txt
+++ b/doc/source/whatsnew/v0.24.0.txt
@@ -500,7 +500,7 @@ Datetimelike
 Timedelta
 ^^^^^^^^^
 
--
+- Fixed bug where :meth:`DataFrame.astype` could not convert timedelta strings (:issue:`#22100`)
 -
 -
 
diff --git a/pandas/core/dtypes/cast.py b/pandas/core/dtypes/cast.py
index e369679d2146f..3f6f0bc4a8ff1 100644
--- a/pandas/core/dtypes/cast.py
+++ b/pandas/core/dtypes/cast.py
@@ -712,7 +712,8 @@ def astype_nansafe(arr, dtype, copy=True):
     elif is_object_dtype(arr):
 
         # work around NumPy brokenness, #1987
-        if np.issubdtype(dtype.type, np.integer):
+        is_time = is_timedelta64_dtype(dtype)
+        if np.issubdtype(dtype.type, np.integer) and is_time:
             return lib.astype_intsafe(arr.ravel(), dtype).reshape(arr.shape)
 
         # if we have a datetime/timedelta array of objects
diff --git a/pandas/tests/dtypes/test_cast.py b/pandas/tests/dtypes/test_cast.py
index 0d6382424ccf5..8c8e6c9a4c242 100644
--- a/pandas/tests/dtypes/test_cast.py
+++ b/pandas/tests/dtypes/test_cast.py
@@ -24,7 +24,8 @@
     find_common_type,
     construct_1d_object_array_from_listlike,
     construct_1d_ndarray_preserving_na,
-    construct_1d_arraylike_from_scalar)
+    construct_1d_arraylike_from_scalar,
+    astype_nansafe)
 from pandas.core.dtypes.dtypes import (
     CategoricalDtype,
     DatetimeTZDtype,
@@ -456,3 +457,13 @@ def test_cast_1d_arraylike_from_scalar_categorical(self):
 def test_construct_1d_ndarray_preserving_na(values, dtype, expected):
     result = construct_1d_ndarray_preserving_na(values, dtype=dtype)
     tm.assert_numpy_array_equal(result, expected)
+
+
+@pytest.mark.parametrize('arr, dtype, expected', [
+    (np.array(['0:0:1'], dtype='O'), 'm8[ns]', 'm8[ns]'),
+    (np.array(['0:0:1'], dtype='O'), 'm8', 'float64'),
+])
+def test_astype_nansafe(arr, dtype, expected):
+    # GH #22100
+    result = astype_nansafe(arr, dtype)
+    assert is_dtype_equal(result.dtype, expected)