To enable testing and performance evaluation of new connected and autonomous driving functions, it is important to characterize packet losses caused by degradation in vehicular (V2X) communication channels. In this paper we suggest an approach to constructing packet loss models based on the socalled Pseudo-Markov chains (PMC). The PMC based model needs only short training sequences, has low computational complexity, and yet provides more precise approximations than known techniques. We show how to learn PMC models from either empirical records of packet receptions, or from analytical models of fluctuations in the received signal strength. In particular, we validate our approach by applying it on (i) V2X packet reception data collected from an active safety test run, which used the LTE network of the AstaZero automotive testing site in Sweden, and (ii) variants of the Rician fading channel models corresponding to two models of correlations of packet losses. We also show that initializing the Baum-Welch algorithm with a second order PMC model leads to a high accuracy model. © 2019 IEEE.