OBJECTIVES: To identify predictive factors of inpatient drug costs in a 500-bed mother-child teaching hospital. METHODS: All hospitalisations in 2000/2001 and 2001/2002 were evaluated. Categorical variables included were major category of diagnosis (MCD) (n=25), severity index (n=4), risk index (n=4) and patient care units (n=41). Continuous variables included were patient weight (kg), level of intensity of resources utilised (LIRU) and total inpatient drug costs/patient-year. Outliers were excluded: inpatient drug costs/patient-year greater than 5000 $CDN, LIRU > 50 and MCD with less than 10 patients per fiscal year. MCD were analysed as serial dichomotical variables. Data were extracted from the admission and the pharmacy software system. RESULTS: Analysis were based on a cohort of 8479patients in 2000/2001 and 7355 patients in 2001/2002. Cost was divided by patient's body weight and log-transformed. A stepwise block multiple regression was processed in two blocks: a first block included LIRU, severity index and risk index and a second block added relevant MCD. Cumulative R2 were respectively 29.3 and 33.5 for both fiscal year. The standardised weight of the variables included for both years were similar. Individual R2 were 15.7 and 19.4 for LIRU, 4.7 and 3.3 for severity index, 0.1and 0.2 for risk index and 8.9 and 10.6 for a selection of relevant MCD. A third of total inpatient drug costs/patient-year can be explained by level of intensity of resources utilised, some major category of diagnosis, severity index and risk index. CONCLUSIONS: There are limited information published on predictive factors of inpatient drug costs/patient-year in hospitals. Further analyses are required to build a useful and stronger model for planning and benchmarking drugs costs in hospitals.