Analyses of variance were conducted using two separate random sets of data (25 158 and 24 084 records) on first-lactation, milk-production records of progeny of 69 British Friesian-Holstein proven sires in order to examine the criteria of seasonal grouping for a sire-evaluation model fitting herd-year-season as a fixed effect. Five different seasonal groupings were chosen each comprising two or three seasons with equal or unequal number of months and with different trends in production within season.
The smallest herd-year × month of calving within-season interaction, smallest residual variance, largest effective number of daughters and smallest among-months, within-season variation were considered to be the appropriate criteria of seasonal grouping. None of the groupings examined satisfied all criteria. However, a small herd-year × month within season interaction and all small within-season variation would be desirable if milk records were pre-corrected for the effect of month of calving in sire evaluation, and the effective number of daughters should be large. Given this the grouping currently used in Britain (three seasons: December to March, April to July and August to November) was considered to be appropriate, although a grouping with two seasons: February to July and August to January gave similar results. The regional differences in production were observed to be large, but the region × month-of-calving interaction was small.