Collision risk modelling of birds at wind turbines typically requires vantage point (VP) data to quantify bird flight activity. The number of VP observation hours required to provide such data, and the associated error in predicted collision rate, have not been formally assessed. Using the Band model and a randomization procedure, we examine the sensitivity of collision rate predictions for the White-tailed Eagle Haliaeetus albicilla to varying hours of input data on flight activity. Variability in collision rate decreased with increasing number of observation hours. However, at the asymptote in variability (about 62 observation hours) there was still considerable variability in predicted collision rate. VP watches are likely to be inherently variable, and collision rate predictions should assess the potential error associated with such results.