diff --git a/t/00-load.t b/t/000-load.t
similarity index 100%
rename from t/00-load.t
rename to t/000-load.t
diff --git a/t/100-new-and-destroy.t b/t/010-new-and-destroy.t
similarity index 100%
rename from t/100-new-and-destroy.t
rename to t/010-new-and-destroy.t
diff --git a/t/101-include.t b/t/011-include.t
similarity index 100%
rename from t/101-include.t
rename to t/011-include.t
diff --git a/t/102-define.t b/t/012-define.t
similarity index 100%
rename from t/102-define.t
rename to t/012-define.t
diff --git a/t/103-code-setters.t b/t/013-code-setters.t
similarity index 100%
rename from t/103-code-setters.t
rename to t/013-code-setters.t
diff --git a/t/README b/t/README
new file mode 100644
index 0000000..ccd2a41
--- /dev/null
+++ b/t/README
@@ -0,0 +1,23 @@
+The test files are ordered thus:
+
+0** - basic and non-compile-based tests
+1** - compile-based tests
+2** - Perl package tests
+3** - other package tests
+4** - tests of other functionality
+
+0-level tests do not require any sort of compilation or interaction with the
+compiled code. They look like normal Perl tests.
+
+100-level tests require interaction with compiled code, but cannot invoke any
+packages. Since TCC::Perl is a package, this means that the only interaction
+is via printing to stdout. As such, all 100-level tests invoke code that
+communicates via C printf's, and use inc::Capture to get the results.
+
+200-level tests focus exclusively on the TCC::Perl package.
+
+300-level tests focus on the other included packages. They can interact with
+the perl interpreter via TCC::Perl, since that will already have been
+tested by the time these run.
+
+400-level tests are for other submodules of the library.