Seventy-seven per cent of 235 newly diagnosed and untreated patients with diabetes mellitus were polysymptomatic at their first visit to hospital. The larger the number of typical symptoms, the greater was the duration of the longest lasting one (p less than 0.001). Symptom duration was positively correlated with pre-treatment fasting plasma glucose (p less than 0.05), but this was more strongly linked to the number of symptoms (p less than 0.001). In those patients diagnosed because of symptoms typical of diabetes mellitus, mean fasting glucose was higher than when symptoms were elicited retrospectively in patients in whom glycosuria had been found unexpectedly. Weight loss at diagnosis (from 'highest remembered') was positively correlated with fasting plasma glucose (p less than 0.001). One hundred and thirty-two patients had fundal colour photography 7 or 8 years after diagnosis. The mean pre-diagnosis duration of symptoms was 2 +/- S.D.17 months in 80 without retinopathy compared to 8 +/- 6 months for 52 with retinopathy (p less than 0.001). The severity of retinopathy was also positively associated with symptom duration, as was failure to attend for these ocular reviews. Prolonged symptoms pre-diagnosis were associated with increased risk of death during the first 5 years post-diagnosis among those with a normal electrocardiogram (ECG) initially (p less than 0.003), but was unrelated to ECG abnormality then or during the next 5 years. These findings accord with (but offer no proof of) the inherently plausible expectation that both morbidity and mortality would be lessened by earlier diagnosis of non-insulin-dependent diabetes mellitus.