The variability of macroinvertebrate assemblages was investigated at 27 upland reference sites in the south-western Cape, South Africa. Multivariate analyses showed that sites did not group on the basis of geomorphological zonation, i.e. mountain stream and foothill-cobble bed. When separate analyses were undertaken for mountain stream (n = 21) and foothill-cobble bed sites (n = 6), assemblages formed three and two groups, respectively. Similarity amongst groups ranged from 47% to 52%, while within-group similarity was between 54% and 67%. Environmental variables shown to contribute to this variability included distance from source, cation ratio ([Na + ]+[K + ]/([Na + ]+[K + ]+[Ca 2+ ]+[Mg 2+ ]), pH, longitude and stream width. Whilst overall variability in the metrics of the biotic index, SASS (South African Scoring System), is high at reference sites, the interpretation of monitoring-site data using biological bands derived from a range of reference sites, ensured that variability was taken into account and that detection of disturbance at a monitoring site was not impeded. A biological banding system has been developed for upland sites in the south-western Cape, together with a list of reference or expected SASS-taxa. This list includes details pertaining to seasonality and biotope preferences. The ability to define reference conditions that take intrinsic variability amongst reference sites into account is important for the accurate interpretation of bioassessment data.