Velocities inside avalanches have been calculated for many years by calculating the cross-correlation between opto-electronic sensors using a method pioneered by Nishimura et al. (1987) and Dent et al. (1998). Their approach has been widely adopted but there has been little discussion of the optimal design of such instruments and the best analysis techniques. This paper discusses some of the different sources of error that arise and how these can be mitigated. A statistical framework that describes such instruments is developed and used to quantify the errors.