diff --git a/pyperformance/_benchmark.py b/pyperformance/_benchmark.py
@@ -13,6 +13,7 @@
 import sys
 
 import pyperf
+from packaging.specifiers import SpecifierSet
 
 from . import _utils, _benchmark_metadata
 
@@ -164,9 +165,12 @@ def runscript(self):
     def extra_opts(self):
         return self._get_metadata_value('extra_opts', ())
 
+    @property
+    def python(self):
+        return SpecifierSet(self._get_metadata_value("python", ""))
+
     # Other metadata keys:
     # * base
-    # * python
     # * dependencies
     # * requirements
 

diff --git a/pyperformance/cli.py b/pyperformance/cli.py
@@ -241,11 +241,15 @@ def parse_entry(o, s):
 
     # Get the selections.
     selected = []
+    this_python_version = ".".join(map(str, sys.version_info[:3]))
     for bench in _benchmark_selections.iter_selections(manifest, parsed_infos):
         if isinstance(bench, str):
             logging.warning(f"no benchmark named {bench!r}")
             continue
-        selected.append(bench)
+        # Filter out any benchmarks that can't be run on the Python version we're running
+        if this_python_version in bench.python:
+            selected.append(bench)
+
     return selected
 
 

diff --git a/pyperformance/tests/data/bm_local_wheel/pyproject.toml b/pyperformance/tests/data/bm_local_wheel/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "pyperformance_bm_local_wheel"
-requires-python = ">=3.8"
+requires-python = ">=3.7"
 dependencies = ["pyperf"]
 urls = {repository = "https://github.com/python/pyperformance"}
 version = "1.0"