Knemometry is an accurate and non-invasive method of quantifying lower leg length changes. It reveals multiple fluctuations in leg length velocity over a short period of measurement, and on the basis of this it has been proposed that short-term growth is saltatory rather than a continuous phenomenon. The technical error (TE) of the technique which is generally employed, and which is subject to observer bias, ranges from 0.09 to 0.16 mm. This study was undertaken to compare the original method (OM) to a modified technique which involved measuring from a baseline value of which the operator was not aware; this technique is referred to as the random zero method (RM). Over a period of 10 months, 58 subjects were measured on 413 occasions. Overall, median TE in the RM group at 0.15 mm (P5-0, P95-0.65) was higher than the median TE in the OM group at 0.11 mm (P5-0, P95-0.37). However, the median TE over the last 3 months of 0.15 mm (P5-0.05, P95-0.87) was lower than the TE in the preceding 4 months of 0.20 mm (P5-0, P95-0.55) (WSR, p = 0.04) pointing towards the presence of an operator learning curve. The random zero method is a simple modification of the original method. It reduces observer bias but leads to a higher TE, which could explain some of the fluctuations seen between frequent knemometric measurements. Some knowledge of the length of the training period is important in the design of new studies involving knemometry; our data suggest that there should be a learning period of about 4 months if knemometry is performed as often as quoted above.