RésuméPour des caractères répétés (production laitière, prolificité...), l'indice de sélection (1 . ) Example of the selection on prolificacy in pigsThe selection index (I.) which is classically used for repeated characteristics (e.g. milk production, prolificacy) is based on an average of n measurements and depends only on the coefficients of heritability (h z ) and repeatability (r) of the trait concerned. As an alternative, a combined index (I . ) i-suggested in which each observation is weighted by a coefficient calculated according to H AZEL ' S theory ; this takes into account genetic correlations (r.) between repeated performances. For a desired breeding goal (H), this index is compared to the conventional one by varying the number of observations made (1 to 5), the values of h 2 (0.1 to 0.6) and r G (0 to 1 in steps 0.1). For a given value of these parameters, the accuracy (correlation between H and I) and the efficiency (expected genetic improvement on H by selecting on I m and I c respectively) of I m and 1, are almost the same. The greatest loss of accuracy of 1 m relative to 1 . is 9 p. 100 ; the difference being more marked as r G is lower and n greater. The accuracy and efficiency of both I m and 1. are improved with increasing n and this is amplified by increasing r o and decreasing h 2 values. If the hypothesis of r! = 1 is wrong, the accuracy and the efficiency of selection may be overestimated ; this overestimation is all the larger as the number of measures is higher ; it also varies with heritability. For example, in the case of selection on prolificacy in pigs (h 2 = .10), overestimation of the expected genetic progress may be 300 p. 100. This character is discussed according to estimations of r G made in the literature ; these estimations are variable but lower than 1 and this may partly explain the discrepancies between theoretical expectations and the actual results obtained in selection experiments.