We study the complexity and propose an algorithm for the problem of determining, given p vectors of {-1, 1}ⁿ, all linear combinations of them which are also in {-1, 1}ⁿ. Computational results are reported. This problem corresponds to the detection of spurious states in Neural Networks.