Urinary growth hormone was measured in 54 children with short stature who had growth hormone deficiency that was initially diagnosed pharmacologically (arginine and L-dopa) and physiologically (mean growth hormone concentration during sleep evaluated twice). Based on the growth hormone response to pharmacological tests the subjects were subdivided into three groups: group A, 20 subjects with normal response (peak concentration > 8 micrograms/l); group B, 20 subjects with response between 4 and 8 micrograms/l; and group C, 14 subjects with response < 4 micrograms/l. In group A four subjects had an abnormally low nocturnal mean growth hormone concentration (< or = 3.3 micrograms/l). In group C seven subjects had multiple pituitary hormone deficiency and abnormal magnetic resonance imaging. All subjects had urine collected from 8.00 pm to 8.00 am for 4-5 consecutive nights. A positive correlation was found between serum nocturnal mean growth hormone values and urinary growth hormone in all subjects. Mean (SD) concentrations of urinary growth hormone were similar in groups A (18.0 (9.5) ng/g creatinine) and B (13.6 (5.9) ng/g creatinine), but significantly higher than that of group C (3.4 (3.7) ng/g creatinine). Considering as abnormal urinary growth hormones below the lower limit of the range in group A, specificity and sensitivity of urinary growth hormone was 100% and 35% respectively. Sensitivity for groups B and C were 5% and 78% respectively. When considering only the subjects of group C with pathological magnetic resonance findings, sensitivity increased to 100%. In the four subjects of group A with mean growth hormone concentration < or = 3.3 micrograms/l, specificity decreased to 80%. It is concluded that urinary growth hormone assay is characterised by a sensitivity too low to be regarded as improving the traditional diagnostic approach to define growth hormone deficiency, unless it is used to identify subjects with the most severe deficiencies.