We study the recognition properties of a modification of the Willshaw
associative memory with a Boating threshold analytically for the asymp
totic case of large networks and single-step retrieval. The number of
spurious stable states and the size of their attraction basins are cal
culated. Comparison is made with simulations for single- and multistep
dynamics. The appropriateness of the recognition procedure based on t
he remanent overlap is demonstrated.