Milk recording data collected in 2,128 dairy herds in England and Wales between 2004 and 2006 were used to predict the calving to conception intervals. The average cumulative milk production was 8,200 kg. Conception (or not) within 5 intervals measured in days (20 to 60 d, 61 to 81 d, 82 to 102 d, 103 to 123 d, 124 to 144 d) was modeled as a function of milk yields and milk constituents at the start of lactation using multilevel discrete-time survival models. Milk yield, weight and percentage of fat, protein, and lactose, and somatic cell counts on the first 2 test-days of lactation were corrected for either stage of lactation alone or stage of lactation and time of year. Five hundred and 1,628 herds, respectively, were used for parameter estimation and cross-validation. Covariates were retained in the final model if their coefficient was at least twice its standard error and their inclusion resulted in a decrease in the deviance. Overall, 73% of cows recalved. The percentage of cows that had conceived by d 20, 61, 82, 103, 124, and 145 were 0.5, 7.3, 17.9, 29.3, 38.7, and 46, respectively. The probability of conception before 145 d in milk increased with lower milk production on the second test-day, higher percentage of protein on the second test-day, and higher percentage of lactose on the first test-day. Positive associations were of a limited magnitude but nonetheless significant with the percentage of protein on the first test-day, the percentage of butterfat on the first test-day, and somatic cell count on both test-days. The model predicted the probability of conception in the cross validation data set very well. Despite the common use of fat to protein ratio as a measure of energy balance, this parameter exhibited wide variation with stage of lactation and time of the year and had a much-reduced ability to predict an early conception compared with other combinations of milk quantity and constituents.