Chaotic dynamics produced by optical delay systems have interesting applications in telecommunications. Optical chaos can be used to transmit secretly, in real-time, a message between an emitter and a receiver. The noise-like appearance of chaos is used to conceal the message, and the synchronization of the receiver with the chaotic emitter is used to decode the message. This work focuses on the study of two crucial topics in the field of chaotic optical communications. The first topic is the synchronization of chaotic external-cavity laser diodes, which are among the most promising chaotic emitters for secure communications. It is shown that, for edge-emitting lasers, two drastically different synchronization regimes are possible. The regimes differ in terms of the delay time in the synchronization and in terms of the robustness of the synchronization with respect to parameter mismatches between the emitter and the receiver. In vertical-cavity surface-emitting lasers, the two linearly-polarized components of the electric field also exhibit isochronous and anticipating synchronization when the coupling between the lasers is isotropic. When the coupling is polarized, the linearly-polarized component that is parallel to the injected polarization tends to synchronize isochronously with the injected optical field, while the other component tends to be suppressed, but it can also be antisynchronized. The second topic is the analysis of time series produced by optical chaotic emitters subjected to a delayed feedback. First, we verify with experimental data that chaos produced by optical delay systems is highly complex. This high complexity is demonstrated by estimating chaos dimension and entropy from experimental time series and from models of optical delay systems. Second, by analyzing chaotic time series, it is shown that the value of the delay of a single-delay system can always be identified, independently of the type of system used and of its complexity. Unfortunately, an eavesdropper can use this information on the delay value to break the cryptosystem. We propose a new cryptosystem with two delayed feedback loops that increases the difficulty of the delay identification problem.