It is known that the thermal effect (thermal inertia) in differential scanning calorimetry causes significant error of measuring the martensite finish temperature (Mf) in shape memory alloy, while the start temperature (Ms) is virtually unaffected. This article shows that the error can be avoided by accounting for the thermal effect quantitatively by mathematical modeling, if the kinetics of the martensitic transformation is properly prescribed. In common with two representative shape memory alloys, Cu-Al-Ni and Ni-Ti alloys, exponential decay is appropriate for expressing the kinetics. The model analysis is extended to the two methods of extrapolation which aims at excluding the thermal effect from DSC data. One is the extrapolation of the cooling rate to zero, and the other is that of the mass of sample to zero. It is shown that both extrapolations construct a temperature between the Ms and Mf. Typically, the temperature is below the Ms by one-third of the interval between the two temperatures.