Stimulus response latency is the delay between stimulus onset and the evoked modulation in neural activity. A common technique to estimate latencies involves binning the spike arrival times to form a peri-stimulus histogram. This histogram is smoothed using a fixed bandwidth. The estimated latency is the first time following stimulus onset in which the smoothed histogram exceeds the midpoint between the minimum and maximum of the smoothed histogram. We demonstrate that the choice of smoothing bandwidth is critical to the accuracy of this latency estimation technique. We suggest a bootstrap resampling technique for bandwidth selection which results in a robust latency estimate.