In modern power systems, frequency measurements are employed in many monitoring and control applications. In this context, Phasor Measurement Units (PMUs) may represent a promising solution in view of a distributed and synchronized measurement infrastructure. The IEEE Std C37.118.1 defines the performance requirements in terms of estimation accuracy and reporting latency, but does not introduce any constraint or guideline with respect to unbalanced or fault conditions. In such operating conditions, the lack of a clear and unanimous definition of power system frequency produces a large variety of alternative estimation approaches and measurand definition. In this paper, we investigate how the adopted definition might influence the measurement outcome with specific reference to the frequency estimation during fault conditions. By making reference to different fault configurations, we compare the frequency estimates and investigate which approach guarantees a more reliable representation of the phenomenon under analysis.