In robotic milking there is always a slight chance of failure to attach the milking cluster. Attachment failure is most likely for cows whose udder conformation is less convenient for robot attachment. In general, after milking failure cows try to revisit the milking robot if they are not sent to a separate area. Since it is difficult to estimate the effect of milking failure on such a cow and her welfare in conditions of robotic milking, a specific 16-day trial was conducted on 12 cows. These cows were milked in a milking parlour with six milking stalls. Each afternoon milking, three cows were not milked. All the cows were closely observed in the cubicle house for 1 h after the afternoon milking. Thereafter, all cows were brought to the milking parlour the third time and the three unmilked cows were milked. In total, each cow was observed 12 times after milking and four times after omitted milking. The following behavioural traits were registered: time budget for the 1 h, occurrence and time until eating, drinking, lying, urination and defecation, and aggressive interactions. Milking order was defined on the basis of how often a cow came to the milking parlour in the first batch of six cows. Moreover, the data related to the milk yield and the use of the automatic feeding installation with the complete diet were analysed. After omitted milking, only the cows from the first batch stood longer in cubicles (14.2 min of 1 h) and lay less (5.4 min of 1 h) than milked cows of the same batch (respectively 7.0 min and 16.3 min for standing and lying in cubicles) (P < 0.01). After omitted milking, cows urinated earlier and more frequently (64.5%) than milked cows (36.3%) (P < 0.002) (both batches). There were no statistically significant differences in eating time and feed intake after milking and omitted milking. Milk yield per cow averaged 24.9 kg during days with omitted (delayed by 1 h) milking and 25.3 kg during the days without omitted milking (P < 0.05). It was concluded that cows show some signs of discomfort after omitted milking (urination); this discomfort seemed to be greater in cows coming earlier to the milking parlour (afterwards they preferred to stand rather than to lie). The 60% of cases of milk leakage found after omitted milking indicates that failed cluster attachment can be accompanied by an extra risk factor for the occurrence of mastitis. However, omitted milking as a treatment did not influence feeding and aggressive behaviour or milking order when unmilked cows were brought to the milking parlour the third time together with the milked cows. Our methods and results can be useful for estimating the effects of robot milking failures on a cow. Future studies should pay particular attention to high-yielding cows and to longer periods of delayed milking. (C) 2000 Elsevier Science B.V.