A phenomenological model is developed to analyze temperature fluctuations in experiments that simulate various liquid-metal fast breeder reactor accident scenarios. Simplified mass, momentum, and energy equations are used to provide a unified system for analysis. Predictions of temperature autocovariance are compared with those observed during a representative in-pile sodium loop experiment. Good agreement is demonstrated both spatially and temporally. General results derived from the analysis of a variety of experiments are summarized and discussed with relation to reactor monitoring.