The Tower of Hanoi (ToH) task was given to 238 children aged from 7 to 15 years, and 20 adults. Individual variation within an age band was substantial. ToH score did not correlate significantly with Verbal IQ, nor with ability to inhibit a prepotent response. We readministered the ToH to 45 children after 30 to 40 days. The test-retest correlation of ·5 is low in relation to accepted psychometric standards, though at least as high as reliability of the related Tower of London (ToL) in adults. The reasons for low reliability remain unclear; task novelty did not seem to be involved, as children did not improve on retest. We conclude that it is not safe to use this test to index integrity or maturation of underlying neurological systems in children. We compared our results with three published studies using the ToL with children, and found similar levels of performance on problems involving the same number of moves. Another study using automated ToL obtained much poorer scores, suggesting that computerised presentation may impair children's performance.