diff --git a/Documentation/Books/AQL/Examples/Join.md b/Documentation/Books/AQL/Examples/Join.md
index e7f6543dbf..a4cd13da29 100644
--- a/Documentation/Books/AQL/Examples/Join.md
+++ b/Documentation/Books/AQL/Examples/Join.md
@@ -220,6 +220,16 @@ So, for each user we pick the list of their friends and count them. The ones whe
 count equals zero are the lonely people. Using *RETURN 1* in the subquery
 saves even more precious CPU cycles and gives the optimizer more alternatives.
 
+### Index usage
+
+Especially on joins you should [make sure indices can be used to speed up your query.](../../ExecutionAndPerformance/ExplainingQueries.md)
+Please note that sparse indices don't qualify for joins:
+
+In joins you typically would also want to join documents not containing the property 
+you join with. However sparse indices don't contain references to documents that 
+don't contain the indexed attributes - thus they would be missing from the join operation.
+For that reason you should provide non-sparse indices. 
+
 ### Pitfalls
 
 Since we're free of schemata, there is by default no way to tell the format of the