OBJECTIVES: To investigate the cost implication of unacceptable refill-based adherence rates of antihypertensive treatment in the South African private health care environment.  METHODS: A non-experimental, retrospective drug utilisation study was conducted on medicine claims data of a pharmacy benefit management company for hypertension patients for a four year period (January 1, 2005 – December 31, 2008). The refill-based adherence rate was calculated per trade name for all antihypertensive medicine items that were prescribed more than once to 171,815 patients. The following equation was used: Refill-adherence rate = (total number of days of antihypertensive medicine items supplied – number of days supplied at last refill) / (date last claimed – date first claimed). [RSA Rand (R)/$US = 6.38112 (2005); 6.78812 (2006); 7.06926 (2007) and 8.27505 (2008)]. RESULTS: An average refill-based adherence rate of 82.8% ± 125.9 was obtained for 376,954 antihypertensive medicine items. Approximately 54.5% of antihypertensive medicine items (n = 205,476) had acceptable refill-based adherence rates between 80% and 120%, 42.2% (n = 159153) had unacceptably low refill-based adherence rates (less than 80%) and 3.3% (n = 12325) had unacceptably high rates (more than 120%). Those with unacceptably low or unacceptably high refill-based adherence rates accounted for respectively 24.3% (R109,275,001) and 1.3% (R5 958,593.70) of the total cost (N = R448,883,473) of all antihypertensive medicine items (n = 376,954). The following pharmacological groups of antihypertensive drugs had on average unacceptable refill-based adherence rates: diuretics (n = 114 623, 74.6% ± 131.7); central acting sympathetic nervous system inhibitors (n = 5 094; 79.9% ± 112.6); sympathetic nervous blockers (n = 3 090; 76.3% ± 87.5). CONCLUSIONS: Since pharmacy records are generally computerised, the prospect of using a simple algorithm to assess compliance based on refill patterns is enticing, particularly for monitoring the progress of patients with chronic diseases.