The objectives of this work were 1) to compare the goodness-of-fit of empirical models of the lactation curve and 2) to analyze the factors affecting the shapes of the lactation curves, the parameters describing them, and the overall milk yield of Latxa dairy sheep. A total of 14,699 records from 2711 ewes, collected during three consecutive years (1995 to 1997) by the milk recording program of the Latxa ewe of the Basque Country (Spain), were used. Six mathematical models and three fitting procedures were compared. The estimation of model parameters by nonlinear fitting procedures was superior to that by linear regression methods. A nonlinear variable decay model fitted the data better than the other models, as judged by lower mean square prediction error, residual sums of squares, and a lack of first-order positive autocorrelation as assessed by the Durbin Watson coefficient. The effects of the flock, flock-year interaction, month of lambing, length of lactation nested within month of lambing, parity, and number of live lambs born had significant effects on the parameters of the model and the total milk yield (P < 0.01). The prediction of milk yield from the selected model was similar to the estimates obtained with the Fleischmann method currently used by the national breeding program for the Latxa breed.