The end-to-end latency of interactive systems is well known to degrade user's performance. Touch systems exhibit notable amount of latencies, but it is seldom characterized, probably because latency estimation is a difficult and time consuming undertaking. In this paper, we introduce two novel approaches to estimate the latency of touch systems. Both approaches require an operator to slide a finger on the touch surface, and provide automatic processing of the recorded data. The High Accuracy (HA) approach requires an external camera and careful calibration, but provides a large sample set of accurate latency estimations. The Low Overhead (LO) approach, while not offering as much accuracy as the HA approach, does not require any additional equipment and is implemented in a few lines of code. In a set of experiments, we show that the HA approach can generate a highly detailed picture of the latency distribution of the system, and that the LO approach provides average latency estimates no further than 4~ms from the HA estimate.