The ‘unitary event’ analysis method was designed to analyze multiple parallel spike trains for correlated activity. The null-hypothesis assumes Poissonian spike train statistics, however experimental data may fail to be consistent with this assumption. Here we present a non-parametrical significance test that considers the original spike train structure of experimental data. The significance of coincident events observed in simultaneously recorded spike trains is estimated on the basis of the same spike trains, however recombined to sets of non-corresponding trials (‘trial shuffling’). Resampling from this set provides the distribution for coincidences reflecting the null-hypothesis of independence.