We present an analysis of the spike response of a retinal ganglion cell ensemble. The retina of a turtle was stimulated in vitro by moving light patterns. Its non-steady motion was specified by two features: changes of direction and changes of speed. The spike response of a ganglion cell population was recorded extracellularly with a multielectrode array and responding neurons were identified through spike sorting. Restricting further analysis to a time window of greatest firing activity, we selected a subset of cells with reliable firing patterns, excluding cells that were not selective to the stimulus. The reliability of a firing pattern was assessed on the single cell level in terms of two measures: temporal precision (jitter) of the first spike and the fraction of trials in which a spike was generated. We then condensed the spike response of the extracted group by merging the multivariate spike trains into a single spike train. Finally, we compared different coding hypotheses that are based on the timing of the first and the second spike of the population or the spike count in the preselected time window. We found that the second spike of the population significantly increases the classification efficiency beyond that of the first spike. Moreover, the combination of first plus second spike is comparably efficient as the combination of the first spike plus the spike count but allows for a classification that is much faster.