We present in this paper a new method of determining the nucleation probability J(T) of a supercooled liquid dispersed within an emulsion, based on the analysis of the thermogram obtained during a regular cooling. The thermal behaviour of the sample during the cooling is modelled in order to calculate the thermogram and so to compare it with the experimental thermogram. A genetic algorithm (GA) is then used to identify the parameters of the J(T) function by minimizing a least squares objective function comparing calculated and measured thermograms. This identification procedure is used to determine the crystallization probability of two samples of same composition but of different manufacture. The results are compared with each other as well as with the results obtained through a traditional calorimetric method. These various tests show that GA associated with calorimetry is a useful and easy tool for the determination of the nucleation probability J(T). (c) 2006 Elsevier B.V. All rights reserved.