To evaluate the within-subject variability of repeated task-induced blood flow velocity (BFV) change, we measured the haemodynamics of the anterior and middle cerebral arteries of the same volunteer on 20 consecutive working days during the performance of mental arithmetic with transcranial Doppler ultrasonography. BFV changes with the arithmetic condition were compared against a resting and a counting condition. Absolute BFV data showed equally increased velocities during rest and arithmetic activity as compared to counting. The findings suggest that the resting state might not be the ideal reference paradigm to evaluate task-induced haemodynamic changes. Selecting the counting condition as the more valid baseline measure, we obtained relatively stable lateralisation indices for the MCA with 1 of 10 measurements outside the 95% confidence interval. Reproducibility was poorer in the ACA, with 3 outliers. A strategy that calculates the average of multiple short activation epochs within the same session appears to improve the reliability of the lateralisation index. (E-mail: guy.vingerhoets@rug.ac.be)