An analysis was conducted on milk production data collected from 1954 to 1981 at Turrialba, Costa Rica. The local conditions are humid and tropical without a sharply defined dry season. The data included purebred Central American Milking Criollos, Jerseys, their reciprocal F, crosses and backcrosses. Two data sets were created: one with minimum restrictions intended to include production data from all calvings, and another data set intended to include only lactations with normal let-down without the presence of the calf. Data were analysed by maximum likelihood according to a model that included the fixed effects of year of calving, parity of lactation, breed or cross and the regression on year of birth of the sire's first daughter and random effects of sire, cow and error. Breed or cross differences were significant for milk yield, fat-corrected milk (FCM) yield and lactation length but were not significant for fat concentration.

F1 crosses were superior to both parental breeds for milk and FCM yields for both data sets; there were no significant reciprocal effects. The ¾ Criollo backcross had significantly lower milk and FCM yields and shorter lactations than did the F1 or ¾ Jersey. These results suggest that breeding programmes for this type of environment should backcross F1's to Jersey sires and use criss-crossing from then on. The proportion of heterosis for milk, FCM, butterfat concentration and lactation length was 0·26, 0·28 0·02 and 0·15 for minimum-restriction lactations and 0·21, 0·22, 0·01 and 0·09 for normal lactations. Heritabilities for milk, FCM, fat concentration and length of lactation were 0·44, 0·42, 0·28 and 0·41 for minimum restriction lactations and 0·28, 0·27, 0·35 and 0·08 for normal lactations. The relatively higher values for minimum-restriction records suggest a strong genetic effect on ease or difficulty of milk let-down.