Each of J items has a characteristic Signature which varies in time. At time 0, the value of a Signature and the identity of the corresponding item are known. No further values of Signatures are observed until a later time T greater than 0. At time t a Signature associated with an unknown item is observed. The problem is to estimate the identity of the item whose Signature is observed at time t. The estimation procedure studied is to estimate the identity of the item that is associated with the Signature at time t to be that one which maximizes the posterior probability of being associated with the observed Signature. Univariate and multivariate Gaussian and univariate Cauchy autoregressive processes are considered as models for the Signatures. The robustness of the univariate Gaussian, (respectively Cauchy), procedure when applied to Cauchy, (respectively Gaussian) data is studied. The results suggest that the Gaussian classification procedure is biased towards classifying the Signature observed at time t as being associated with the same item that is associated with the Signature at time 0. The Cauchy procedure is biased towards classifying a Signature observed at time t as being associated with a different item than the one associated with the Signature at time 0.