diff --git a/test/Performance/README.md b/test/Performance/README.md
new file mode 100644
index 000000000..7a23147a9
--- /dev/null
+++ b/test/Performance/README.md
@@ -0,0 +1,54 @@
+Running Performance Tests
+=========================
+
+Pre-Requisites
+--------------
+
+* Python 2.7+ or 3.5+
+* msbuild.exe (must be on `PATH`)
+
+Single Perf Run
+---------------
+
+1. Build the CLI repo to get dotnet.exe, or otherwise source the CLI. For
+   meaningful perf results, be sure to use release mode.
+
+2. `cd <cli_repo_root>/test/Performance`
+
+3. `python run-perftests.py <dotnet_bin> --name <unique_run_name>
+   --xunit-perf-path <x_repo_path>`  
+   where:
+    * `<dotnet_bin>` is the path to the dotnet binary whose perf you want to
+      measure.
+    * `<x_repo_path>` should point either to an non-existent directory, or to
+      the root of a local clone of xunit-performance. If a non-existent
+      directory is specified, the repo will automatically be cloned.
+        - NOTE: You can also set the environment variable
+          `XUNIT_PERFORMANCE_PATH` to avoid having to pass this variable every
+          time.
+
+4. View the `*.csv` / `*.xml` results in the current directory.
+
+Comparison Run
+--------------
+
+In general, follow the same steps as for a single perf run. The following
+additional steps are required:
+
+1. In addition to the dotnet.exe that you're testing, be sure to also build or
+   otherwise source the baseline dotnet.exe. This could be the "stage0" exe, or
+   the exe from the last nightly build, or the exe built from sources prior to
+   changes you made, etc.
+
+2. When invoking `run-perftests.py`, add an additional parameter: `--base
+   <base_bin>`, which points to the baseline dotnet.exe mentioned in step 1.
+
+3. View the `*.html` file generated for the perf comparison analysis.
+
+Debugging Issues
+----------------
+
+The output of commands invoked by `run-perftests` is hidden by default. You can
+see the output after an error by looking in the `logs/run-perftests` directory.
+Alternatively, you can rerun `run-perftests` with `--verbose`, which will print
+all output to the console instead of piping it to log files.