The behaviour of 36 Yorkshire X Netherlands Landrace sows was individually and continuously recorded during eleven, 5-min periods commencing at 10.00 h. The sows were neck tethered in partial stalls in the dry sow accomodation of a 250-sow, commercial herd. The recorded behaviour data were analysed for cycles of repetitive behaviour which appeared to be non-functional. Four classes of stereotyped behaviour were recognized: fixed complete routines, fixed subroutines, transitory and loose. They were performed during 25.8, 12.9, 4.9 and 2.7% of the observation time, respectively. Some aspects of the two fixed-class stereotypes were found to be related to stage of pregnancy, that is, time spent tethered in the current pregnancy. The proportion of observation time that sows performed fixed stereotypes, and the number of fixed stereotypes, increased per sow with increase in stage of pregnancy, although in the former instance, less time was spent performing the stereotypes after day 80 of pregnancy. Considerable variability was recorded in the stereotypes in terms of use of particular elements, duration of element use, etc., suggesting that the classical definition of stereotyped behaviour may only hold in situations where the environment is totally unchanging. The use of a quantitative approach to measure stereotypes in tethered sows provides the basis for investigation of cause and function of these behaviours, and therefore the interpretation of stereotypes in terms of sow welfare.