The geometry plays a critical role in the design and analysis of ad hoc networks, and the distribution of nodes is naturally modeled by a stochastic point process. Due its analytical tractability and "complete spatial randomness" property, the stationary Poisson point process has been the model used in most work so far. In most ad hoc networks, however, the assumptions of independence and no interaction do not hold. There is therefore a need to consider other point process models that better describe the actual network geometry. We present and discuss several such models and their potential applications, paying special attention to the trade-off between accuracy and analytical convenience. In particular, we focus on inhomogeneous Poisson point processes, hard core processes, cluster and lattice processes, and marked point processes.