We examine a specific candidate for temporal coding of information by spike trains, the occurrence of a temporal firing pattern among some number of neurons that repeats more often than expected by chance. Methods for detection of repeating patterns have long been available, but there are no analytic methods for calculating the expected numbers of repeating patterns to enable assignment of significance to the results from the experimental data. The expected numbers can be calculated by Monte-Carlo methods by repeatedly modifying the original data spike trains. Ideally the surrogates produced by such changes should destroy all patterns and cross-correlations but preserve other aspects of the trains such as rate, interval structure etc. We present here a novel variant of the "dither surrogate" (Date et al. 1998) and use surrogates generated by this algorithm to evaluate repeating pattern significance in data recorded in monkey motor cortex during behavior. Although we can demonstrate high statistical significance for the excess repetition of some spike patterns, it is not obvious that this has physiological meaning or that such patterns are used for information transfer.