The aim of the research was to improve a method of estimation of test-score reliability of a single trial. The hypothetical correlation between the first trial and a trial on other occasions assuming no learning or fatigue effects was determined. It was shown that a valuation of the single measure test reliability using an intraclass correlation coefficient may show an understated assessment because trends in the repeated trials caused by combination of fatigue and training effects. A method of elimination of systematic changes in repeated trials based on linear regression was proposed. In a numerical example, elimination of these trends resulted in an improvement of assessment of the test score reliability from a near acceptable level (0.785) as ICC (3,1) to an acceptable level (0.803) as ICC (1,1).