The aim of this study was to examine the reliability of the twitch interpolation technique when used to estimate the true isometric knee extensor muscle strength. This included an examination of whether submaximal activation causes any bias in the estimation of the true muscle strength and an examination of the precision of the method. Twenty healthy subjects completed three contraction series, in which the subjects were told to perform as if their voluntary strength was 60%, 80% or 100% of that determined by a maximal voluntary contraction (MVC). Electrical muscle stimulations were given at each of five different contraction levels in each series. At torque levels above 25% of MVC the relationship between torque and twitch size could be approximated to be linear. The true muscle strength (TMS) could therefore be estimated using linear regression of the twitch-torque relationship to the torque point of no twitch in each of the three series, termed TMS60, TMS80 and TMS100. The TMS80 was slightly lower (P < 0.01), median 94% (IQ range 87-101%) of the TMS100. The TMS60 was median 99% (IQ range 83-125%) (NS) of TMS100, but a few severe outliers were observed. In conclusion, we found the reliability of the method acceptable for many research purposes, if series with estimated central activation of below 40-50% were excluded. The only moderate precision and the slightly lower estimations in subjects applying submaximal does, however, limit its usefulness.