Spatial knowledge, necessary for efficient navigation, comprises route knowledge (memory of the landmarks along a route) and survey knowledge (map-like). Available data on the retention in humans of spatial knowledge show that this does not decline systematically over months or years. Here, two groups of participants elaborated route and survey knowledge during navigation in a complex virtual environment before performing route and survey tasks. Both groups were tested 5 minutes after learning and 3 months later, while one group was also tested 1 week and 1 month later (repeated testing). Performance was similar in both groups on the first testing session, remained stable in the repeated tested group, but decreased in the non-repeated tested group, especially on route tasks. These results are the first to reveal a substantial and selective decline of spatial knowledge, occurring only if there is no possibility of reactivating knowledge along repeated testing.