This article analyses the recently suggested particle approach to filtering time series. We suggest that the algorithm is not robust to outliers for two reasons: The design of the simulators and the use of the discrete support to represent the sequentially updating prior distribution. Here we tackle the first of these problems.