The objectives of the present study were to compare feed selection in tie- vs. free-stall barns and also to verify possible correlations between feed selection and milk composition. Forty multiparous and 20 primiparous lactating Holstein cows were utilized in a crossover design with 21-d periods. Cows were randomly divided into 2 groups; group 1 was housed in a free-stall barn during period 1 and a tie-stall barn during period 2, and vice versa for group 2. In the free-stall barn, 18 extra cows were also present. Animals were fed the same diet once daily in the free-stall barn at 1100 h and twice daily at 1100 and 1500 h in the tie-stall barn to obtain approximately 10% daily refusals in both facilities. Group feed selection in the free-stall barn was measured and compared with the group feed selection in the tie-stall barn, obtained by summing individual feed offered and refused. Feed selection was analyzed including treatment and period in the model. Sequence effect and true error were combined into a single degree of freedom error term. Intake of the longest particles expressed as a percentage of the predicted intake was 73.2% in a tie-stall barn and 63.3% in a free-stall barn. There were no significant correlations between individual feed selection measured in the tie-stall barn and milk composition. Feed selection estimates made with individually fed cows are likely to underestimate average feed selection in a free-stall barn.