The objective of this study was to estimate and model the percentage of protein, fat, lactose, non-fatty solids and protein-fat relationship variation in a 90-day milking period, from 136 test-day milk yield records of 46 Chiapas ewes. Least square means were estimated using a mixed model with repeated measures considering year (2006 to 2007), lactation (2, 3, 4, 5), variety (white, brown and black) and the interaction between lactation and variety. The relationship between days of lactation and daily milk yield (ml) and composition was modeled using random regression techniques. Least square means were 14.2 ± 0.36 kg for milk yield per lactation, 169.12 ± 4.97 ml/ewe/day, 5.49 ± 0.04% for protein, 4.37 ± 0.17% for fat, 4.53 ± 0.03% for lactose, 11.08 ± 0.04% for non-fat solids and 1.56 ± 0.07 for protein-fat relationship. Daily milk yield showed constant decreasing, while milk components presented quadratic trend during milking period. The component percentages of protein, fat, lactose, non-fatty solids and protein: fat relationship remained constant during the first five lactations and varieties showed similarity between milk composition studied traits, except in fat content, where the white variety had the highest proportion and the black variety the lowest, with a difference of 30%, whereas the brown variety was intermediate between these two. The results of the present study show the feasibility of selecting the Chiapas sheep breed for milk production and for a dual-purpose animal (wool-milk) under grazing conditions in the Altos de Chiapas, Mexico. Key words: Protein in sheep milk, fat in sheep milk, lactose in sheep milk, Composition of sheep milk, Chiapas sheep breed, modeling milk composition, random regression.