The introduction of fully automatic milking systems (AMS) on dairy farms can only succeed if cows visit the system more or less voluntarily, at regular intervals. This can be arranged by either forced or free cow traffic. In the case of forced cow traffic, the AMS is the only route from the lying area to the feeding area. An alternative is free cow traffic, where the cows can decide whether to visit the AMS or not. The behaviour in 3 x 20 crossbred Holstein Friesian dairy cows in three experiments (experiment Ia, Ib and II) was studied in a cubicle house with a selection system consisting of a selection stall and a concentrate feeder, simulating an AMS. Cows could obtain 250 g of concentrate in the system once every six hours. Actual milking occurred in a conventional milking parlour twice a day. Experiment Ia consisted of (1) a reference phase with no use of the selection system, (2) free cow traffic, (3) forced cow traffic, and (4) a reference phase. Experiment Ib was a repetition of experiment Ia. Experiment II was divided into (1) a reference phase with no use of the selection system, (2) forced cow traffic with two one-way passages between the feeding and lying areas, (3) forced cow traffic with one one-way passage, (4) free cow traffic with one free passage between the feeding and lying areas, and (5) free cow traffic with two such passages and (6) a reference phase. The number of visits paid to the selection system by individual cows appeared to be consistent in the different phases in each experiment (r = 0.52 in experiment I (P < 0.05); r = 0.68−0.86 (P < 0.01) in experiment II). In experiments Ia and Ib, cows seemed to visit the selection system more often during forced cow traffic than during free cow traffic, but this was because some of the cows did not visit the system at all during free cow traffic. Therefore, the differences were not significant. In experiment II, cows paid the same number of visits to the selection system during different types of cow traffic, but when two passages were open during free cow traffic, the cows visited the system less (P < 0.01). During forced cow traffic, cows spent more time standing on the slatted floor in the feeding area, spent less time standing at the feeding gate and made less journeys from the lying to the feeding area (significant in two of three experiments; P < 0.01). It was concluded that forced cow traffic may improve the frequency of visits to the AMS, but somewhat restricts the cows' behaviour, and may, therefore, be questionable. Free cow traffic could work if cows are previously conditioned to take the route to the AMS.