The effect of reducing the frequency of official milk recording and the number of recorded samples per test-day on the accuracy of predicting daily yield and cumulative 305-day yield was investigated. A control data set consisting of 58 210 primiparous cows with milk test-day records every 4 weeks was used to investigate the influence of reduced milk recording frequencies. The accuracy of prediction of daily yield with one milk sample per test-day was investigated using 41 874 testday records from 683 cows. Results show that five or more test-day records taken at 8-weekly intervals (A8) predicted 305-day yield with a high level of accuracy. Correlations between 305-day yield predicted from 4-weekly recording intervals (A4) and from 8-weekly intervals were 0.99, 0.98 and 0.98 for milk, fat and protein, respectively. The mean error in estimating 305-day yield from the A8 scheme was 6.8 kg (s.d. 191 kg) for milk yield, 0.3 kg (s.d. 10 kg) for fat yield, and −0.3 kg (s.d. 7 kg) for protein yield, compared with the A4 scheme. Milk yield and composition taken during either morning (AM) or evening (PM) milking predicted 24-h yield with a high degree of accuracy. Alternating between AM and PM sampling every 4 weeks predicted 305-day yield with a higher degree of accuracy than either all AM or all PM sampling. Alternate AM-PM recording every 4 weeks and AM + PM recording every 8 weeks produced very similar accuracies in predicting 305-day yield compared with the official AM + PM recording every 4 weeks.