In Automatic Milking Systems (AMS), cows are kept in a free stall barn equipped with one or several milking units (MU). The cows’ activities in the system are called cow traffic. There are three types of cow traffic systems: guided, partly-guided and free. Free systems have no gates, allowing the cows access to the feeding and resting areas at any time. This study pre-sents the results of on-farm research on such free systems, especially with a view to produc-tion results, feeding and management routines, animal health, and housing. The underlying field work was undertaken on nine commercial dairy farms in Sweden and comprised of man-ager (farmer) interviews, own observations at the target-barns, and the acquisition of standard data that were collected and stored by the farms’ AMS software programmes. The data, cov-ering the system responses and interactions of 837 lactating cows in 15 MUs over a period of 30 days were subsequently processed by aid of the statistical pc-programme SAS and Micro-soft Excel. With each of the 15 investigated MUs shared by 40 to 65 cows, the milking frequency varied from 2.50 to 2.84 milkings per cow and day at intervals between 8.4 and 9.6 hours. While there was no relationship between milking frequency and annual milk yield or herd size, the latter clearly affected milking intervals in a positive fashion. The number of non-milking (ex-tra) visits to the MUs varied between 0.71 and 1.51 per day. When cows paid up to two extra visits and five visits in total to the MUs, milkings took place at shorter intervals and were more evenly distributed over the day. However, no additional positive development in milk-ing interval could be seen for more than two extra visits. There was a tendency for a signifi-cant negative correlation (p<0.09) between the number of extra visits per cow and her annual milk yield. There also was a significant negative correlation (p<0.001) between the number of extra visits per cow and her daily milk yield. The share of unsuccessful milkings varied widely, ranging from 1.94 to 12.19 % of all milkings (median 3.17 %). If there were specific times for fetching cows that were late for milking, cows with unsuccessful milkings were also fetched, even though the farmers were not aware of this. The feeding patterns applied on the participating farms were distinctly different from each other and ranged from 2 to 16 times per day. Whereas all nine farmers stated that the cows were fed roughage ad lib, only two actually had feed on the feeding table at all times. On four of the farms, the feeding table was empty every morning. In consequence milking activities were highest after morning feeding and low prior to feeding. Only 56 % of the farmers had an idea how much milk they were feeding for at the feeding table. There was a significant nega-tive correlation (p<0.023) between the maximum allowance of concentrate per cow and day in the MU and the annual milk yield per cow. There was a tendency for a positive correlation (p<0.093) between the maximum concentrate allowance in the MU and the number of extra visits, however without affecting the number of daily milkings. The fetching frequencies were quite low, compared to what has been observed in research studies. Free time in the MU had a large effect on the number of involuntary milkings (fetch-ings). For free time over 11 %, there were 2.1 % involuntary milkings, but for free time less than 11 % there were 5.0 % involuntary milkings. There was a significant positive correlation (p<0.036) between the proportion of involuntary milkings and herd size. The average fetching frequency for the farms with less than 55 cows per MU was 2.3 % (1.0 – 3.8 %) and for the farms with more than 60 cows it was 7.1 % (6.1 – 8.0 %). The average minimum interval for fetching was 13 hours, with variations from 10 to 20 hours. There was a negative correlation (not significant) between the fetching frequency and the minimum interval for fetching. When there were no specific times for fetching, the milking interval was more regular during the day, but somewhat longer than on farms with specific fetching routines. Cleaning of the MU had a strong effect on the number of milkings per hour, and there was a drop in milkings already some time before the actual cleaning. Usually there was a peak in milkings after cleaning. However, if the feeding table was empty prior to cleaning and feeding occurred shortly after cleaning, the peak was levelled off. There were considerable differences in the average milk flow rate, varying from 2.01 to 2.56 kg milk/minute (mean = 2.21). It turned out that herd size in terms of cows/MU was significantly positive correlated to the av-erage milk flow rate in the herd (p<0.013). There were great differences regarding the average bulk milk somatic cell count (BMCC). It varied between 140 000 and 275 000 cells/ml (average 206 000). There was no pattern ob-served, when studying health problems on the investigated farms. Investments in AMS were primarily motivated by the need to provide for more flexibility in daily farm management. Other reasons were the reduction of heavy work related to conventional milking and to be less dependent of extern employees. A technical interest was also mentioned for motivation. Alarms and on-call duty were not a problem for any of the interviewed farmers.