Optical Burst Switching (OBS) is a new paradigm for future all-optical networks. OBS networks are difficult to investigate analytically, so most of the current research is based on simulations. This has created a need for accurate traffic sources. In this paper, we present an overview of different traffic sources and make a distinction between simulating burst assembly and generating bursts (where no information about burst contents is carried). The following types of sources are discussed: emulating assembled traffic using exponentially distributed burst length and interarrival time, emulating assembled traffic with burst length and interarrival time distributions and burst assembly simulation with LRD input traffic. We discuss advantages and disadvantages of each approach, mainly in respect to its efficiency. For example, simulating burst assembly with LRD input traffic, while considered perfect in terms of quality of produced traffic, is also the least effective and requires long simulation times. We show the area of applicability of each type of traffic generator, comparing it to a small number of burst assemblers, or to highly aggregated OBS traffic. We also present two new types of traffic sources: one that can generate traffic with LRD properties in a much more efficient way than full burst assembly simulation, and one that uses conditional burts length and interarrival time distributions. Finally, simulation results are presented to confirm our findings.