Thermoluminescence dosimetry (TLD) for radiotherapy treatment verification is performed in the Prince of Wales Hospital in Sydney for a wide range of applications: (A) to determine the dose in difficult treatment geometries, (B) to record the dose to critical organs, and (C) to monitor special treatments such as total body irradiation (TBI). TLD measurements were performed with the aim to investigate cases where dose prediction is difficult and not as part of a routine verification procedure. We reviewed 1058 reports of TLD performed during the treatment of 502 patients between 1986 and 1991 to evaluate how the TLD results compare with the dose determined by the treatment plan. Reasons for possible discrepancies should be identified. In 19% of all investigated cases a discrepancy of more than 10% was found between expected and measured doses. The discrepancies could be divided into three groups: (1) errors made in the TLD determination or evaluation, such as placement errors of the TLD chips (21% of all discrepancies); (2) mistakes made during the patient set-up, such as insufficient shielding or inadequate patient immobilisation (30%); (3) inadequate treatment planning and dose calculation procedure, such as wrong inverse square law corrections or errors due to limitations of the two-dimensional treatment planning system used (41% of all). In 8% of all discrepancies the reason remained unclear. A number of changes to treatment plans and modalities (e.g. changed scrotal shield, modified bolus) were introduced due to TLD results. The increasing number of TLD requests per year attests to the value of TLD as a treatment verification method in clinical practice.