Clinical images show that T2 contrast allows in certain cases a better discrimination between the healthy and pathological tissues. For instance, in Fig. 1 it can be seen that multiple sclerosis’ lesions are more clearly visible on the images obtained with late echoes than those obtained with the early echoes. On the other hand in vitro studies on perfused organs’ have shown that the T2 decay curve depends on the interval between 180” pulses: The decay curve is mono-exponential for 4-ms time intervals between 180” pulses, and bi-exponential for 0.4-ms time intervals (Fig. 2). For these reasons, we have tried to analyze the T2 decay curve by multi-exponential decomposition in order to improve tissue characterization. To obtain the T2 decay curve, we used the classical CPMG pulse sequence. By adding linear magnetic field gradients to this sequence it is possible to reconstruct a sample image for each echo. The T2 decay curve is constructed by taking the corresponding pixel from the echo image series, for instance, the first point corresponds to the first image, the second point to the second images, etc. Thus we obtain the T2 decay curve of the elementary volume defined by the pixel whose dimensions are the slice thickness (7 mm) and the spatial resolution (2-2.5 mm). It suffices, therefore, to repeat this operation pixel by pixel in order to obtain parametric T2 decay curves over the whole slice. A slice of the head is represented by about 6000 pixels, which means that we need to repeat the decomposition operation 6000 times. We therefore had to find an algorithm sufficiently rapid and not too memory consuming. The algorithm chosen was of the noniterative, Prony (algebraic) type. 3 It was then necessary to verify this method of analysis using first simulated image signals and thereafter simple test objects. Classically, the multi-exponential decomposition is an ill-posed problem. The “solution” is obtained by nonlinear least-squares optimization. With a noniterative algorithm we transform the nonlinear leastsquares optimization into two linear least-squares problems. The decomposition method may be presented in four steps: