This paper assesses the performance of a new simulated maximum likelihood (SML) estimator that is particularly useful for estimating dynamic panel data models with unobserved endogenous state variables. The performance of the estimator is assessed via repeated sampling experiments on a panel data probit model with a time-varying exogenous covariate, lagged endogenous variables and serially correlated errors. The results of a series of a series of experiments that allow for state-dependence, individual random effects and AR(1) serially correlated errors show that the new SML estimator has good small sample properties and is computationally tractable for use with panels of the size that are likely to be encountered in practice.