The fetal behavioural state concept was used to study fetal rest-activity cycles in normal pregnancies at gestational ages of 32 and 38 weeks. In addition, it was investigated if clustering of fetal movements was already present in recordings obtained at 20 weeks. At 20 weeks, 17 periods lasting longer than 3 min were found in which fetal body movements were absent. The mean duration of these periods was 4.0 +/- 0.8 min. On the basis of random scattering of movements on a time axis, it appeared unlikely that these periods of inactivity occurred by chance alone. At 32 and 38 weeks, data on body movements were combined with data on eye movements and the FHR pattern. At 32 weeks, true fetal behavioural states were not found. The average coincidence of 1F to 4F was 58%, while only 23% was to be expected if state parameters had fulfilled state criteria purely by chance. At 38 weeks, coincidence of 1F to 4F had increased to 80% (P less than 0.001). Expected coincidence purely by chance was 30%. True fetal behavioural states were found in 17 out of 35 recordings. For transitions from 1F into 2F, the FHR changed relatively early, i.e. as first or second parameter, while for the reverse transitions it changed relatively late (P less than 0.05). There was no clearly preferred sequence for body and eye movements within transitions.