In this paper, a generalized expression for latent heat has been derived by considering temperature gradients within the sample, and the usual situation in which the specific heat of the sample changes during the transition and specific heats in both low- and high-temperature states are functions of the temperature. For a reversible transition, when the specific heat of a sample is a slowly changing function of the temperature, the errors caused by the temperature gradient within the sample can theoretically be minimized by averaging the latent heats obtained from both the heating and cooling process. It is also pointed out that the latent heat is proportional to the area enclosed by peak curve and the imaginary equivalent non-latent heat baseline.