Hybrid automata offer a framework for the description of systems with both discrete and continuous components, such as digital technology embedded in an analogue environment. Traditional uses of hybrid automata express choice of transitions purely in terms of nondeterminism, abstracting potentially significant information concerning the relative likelihood of certain behaviours. To model such probabilistic information, we present a variant of hybrid automata augmented with discrete probability distributions. We concentrate on restricted subclasses of the model in order to obtain decidable model checking algorithms for properties expressed in probabilistic temporal logics.