This observational study investigated the relationship between balance, mobility and falls in 72 hemiplegic stroke inpatients, with the aim of developing a model for predicting fall risk. Fall history was recorded by interview, balance was assessed using the Berg Balance Scale (BBS) and activities of daily living were evaluated using the Functional Independence Measure (FIM). Variables differing between fallers and non-fallers were identified, and a stepwise regression analysis was performed to identify a combination of variables that effectively predicted fall status. Fallers (occasional and repeat; n = 27) had a shorter time from stroke onset, lower FIM scores on admission and discharge, lower BBS and Mini-Mental State Examination scores, a greater age and longer length of hospital stay compared with non-fallers (all differences were significant). A logistic model for predicting falls showed that BBS at admission was significantly related to falls, with fallers having lower BBS scores at admission (cut-off <or= 29; sensitivity 80%; specificity 78%). These data suggest BBS is a sensitive and specific measure for identifying stroke patients at risk of falling.