One trivial situation in which this can happen is when the tissue used for expression studies is heterogeneous. Different cells express different levels of the gene.
Bimodality can be observed when the system can actually occupy two stable states; i.e. a gene can either have a high expression or a low expression. When you sample the population you would get two peaks. Bistablity (two stable steady states) is a common phenomenon in biological systems and positive feedbacks generally exhibit such behavior. In bistable systems there is also an unstable steady state which lies "between" the two stable states (like a mountain separating two valleys). If the system is in the unstable state, it can fall to either of the two stable states. (See this article for an example). This concept can be extended to multistable systems but they are a little more complex than the simple feedbacks. However they can theoretically exist (I don't know of a biological example yet).
Bimodality/multimodality can be also observed in the absence of a deterministic bistability in the system. This happens because of the expression noise due to stochasticity and is observed in case of transcription bursts (See here).