Abstract. Identification is a useful cryptographic tool. Since zero-knowledge theory appeared [3], several interactive identification schemes have been proposed (in particular Fiat-Shamir [2] and its variants [4,6,5], Schnorr [9]). These identifications are based on number theoretical problems. More recently, new schemes appeared with the peculiarity that they are more efficient from the computational point of view and that their security is based on N P-complete problems: PKP (Permuted Kernels Problem) [10], SD (Syndrome Decoding) [12] and CLE (Constrained Linear Equations) [13]. We present a new N P-complete linear problem which comes from learning machines: the Perceptrons Problem. We have some constraints, m vectors X i of {−1, +1} n , and we want to find a vector V of {−1, +1} n such that X i · V ≥ 0 for all i. Next, we provide some zero-knowledge interactive identification protocols based on this problem, with an evaluation of their security. Eventually, those protocols are well suited for smart card applications.