This paper presents a technique for studies of stability of soil aggregates. The major soils of the State of São Paulo in different conditions of use have been examined. Triplicate samples of 25 g of soil aggregates of 7-4 mm size for each soil were shaken in a Model Wagner shaker at 40 rpm during 1, 2, 4, 8, 16, 32, 64 and 128 minutes period and were later fractioned by the wet sieving method. In general soil aggregates greater than 2 mm decreased gradually in weight with increasing period of treatment: at the some time the small size particles (< 0.2 mm) increased in weight. The intermediate aggregates behaved differently in different soils. Statistical analysis of the classes of aggregates > 2 mm and < 0.2 mm gave the general information needed. In general the experimental error were homogeneous and of small magnitude in relation to the systematic differences existent. The functional relationship between the two variables (time and weight of size particles) was calculated by orthogonal polynomials considering the log time scale. In general polynomials of second degree were sufficient for a good approach, but in some cases polynomials of third degree were necessary. The authors believe that for fertility studies the interval 1-32 minutes is sufficient to separate the differences in aggregabifity. In that interval for the soils considered a first degree polynomial is satisfactory. Intervals of confidence had been calculated for the polynomial Y = A + B'E'1. The values Ai, As, B'i, B's were respectively the inferior and superior limits of the respective interval. Separations due to different use of soils were obtained through such statistics. As the variability between triplicates had been small and the data obtained very consistent, the authors believe that for future studies duplicate determinations would furnish the estimate of error with sufficient precision.