diff --git a/README.md b/README.md
index 8ecf45a..0ab31ff 100644
--- a/README.md
+++ b/README.md
@@ -584,6 +584,48 @@ to the second element in the schema, and succeed:
 
 ```
 
+## Multi-field validation
+
+Validation rules that involve multiple fields can be implemented as
+custom validators. It's recommended to use `All()` to do a two-pass
+validation - the first pass checking the basic structure of the data,
+and only after that, the second pass applying your cross-field
+validator:
+
+```python
+def passwords_must_match(passwords):
+    if passwords['password'] != passwords['password_again']:
+        raise Invalid('passwords must match')
+    return passwords
+
+s=Schema(All(
+    # First "pass" for field types
+    {'password':str, 'password_again':str},
+    # Follow up the first "pass" with your multi-field rules
+    passwords_must_match
+))
+
+# valid
+s({'password':'123', 'password_again':'123'})
+
+# raises MultipleInvalid: passwords must match
+s({'password':'123', 'password_again':'and now for something completely different'})
+
+```
+
+With this structure, your multi-field validator will run with
+pre-validated data from the first "pass" and so will not have to do
+its own type checking on its inputs.
+
+The flipside is that if the first "pass" of validation fails, your
+cross-field validator will not run:
+
+```
+# raises Invalid because password_again is not a string
+# passwords_must_match() will not run because first-pass validation already failed
+s({'password':'123', 'password_again': 1337})
+```
+
 ## Running tests.
 
 Voluptuous is using nosetests: