[en] A candidate technique for detecting incipient blockages in the fuel sub-assemblies of liquid-metal-cooled fast-breeder reactors is the measurement of coolant outlet temperature fluctuations caused by turbulence. A theoretical basis for the method is required, and one model is discussed in this paper. The model requires the paths of individual fluid particles to be traced, allowing their motions to vary randomly, subject to certain constraints of the turbulent motion. It is a Monte Carlo method, already used successfully in predicting the transport of scalar fluctuations. Its advantage is that it allows actual time-dependent temperatures to be computed, and so various methods of analysis (power spectral density, amplitude probability distribution, for example) can be examined. The effect of heat conduction is neglected at present. Computed results show the way in which mean temperature and rms temperature-noise profiles develop in a turbulent flow in a pipe downstream of a plane at which a steady mean profile is maintained. Power spectral densities and amplitude-probability densities are presented, and it is shown how an amplitude-probability density plot distinguishes between a temperature-gradient input profile and a gaussian input profile. Comparisons are given with experimental measurements made in representative heated pin rigs in sodium, with good agreement. Some of the problems associated with extrapolation to real sub-assemblies are discussed. (author)