A small (28 g) mechanical accelerometer has been tested by subjecting it to controlled bench tests consisting of repetitive vertical oscillations on two designs of test rig. The accelerometer's 3-digit display provided a cumulated score with a maximum of 99.9 units. This score was compared with an independent count of the imposed oscillations and found to be linear with time (r = 0.996) and reproducible on retest (coefficient of variation = +/- 1.5%). The sensitivity ranged from 6.2 to 7.4 units/10,000 oscillations. The response was related to the maximal applied acceleration (calculated from the amplitude and frequency of the oscillations on the assumption that they were sinusoidal) and independent of the amplitude and frequency used. The threshold maximal acceleration was less than 2 m s-2 and the response had reached a plateau at 4 m s-2. During field studies the accelerometer was firmly attached over the hip in a waistband where it responded to the vertical accelerations produced by walking. When compared with an independent count of footsteps from a heel-mounted resistance pad the accelerometer score (after calibration) was not significantly different. The mean difference was (0.29 +/- 0.67, S.D.) 10(3) "steps" in a younger group (n = 8, mean age 39 years) and (0.46 +/- 1.08, S.D.) 10(3) "steps" in an older group of women (n = 6, mean age 65 years). Scores of around 10 X 10(3) "steps" can be expected in a day in moderately active young subjects and 40 X 10(3) "steps" in a week in the elderly. Simultaneously recorded scores from both right and left hips wee not significantly different.(ABSTRACT TRUNCATED AT 250 WORDS)