Recently, we have published a method for the calculation of required leaf trajectories to generate optimized intensity modulated x-ray beams by means of dynamic multileaf collimation [Phys. Med. Biol. 43, 1171–1184 (1998)]. For the MM50 Racetrack Microtron it has been demonstrated that the dosimetric accuracy of this method, in combination with the dose calculation algorithm of the Cadplan 3D treatment planning system, is adequate for a clinical application (within 2% or 0.2 cm). Prior to initiating patient treatment with dynamic multileaf collimation (DMLC), tests have been performed to investigate the stability of DMLC fields generated at the MM50, (i) in time, (ii) subject to gantry rotation and (iii) in case of treatment interrupts, e.g., caused by an error detected by the treatment machine. The stability of relative dose profiles, normalized to a reference point in a relatively flat part of the modulated beam profile, was assessed from measurements with an electronic portal imaging device (EPID), with a linear diode array attached to the collimator and with film. The dose in the reference point was monitored using an ionization chamber. Tests were performed for several intensity modulated fields using 10 and 25 MV photon beams. Based on film measurements for sweeping 0.1 cm leaf gaps it was concluded that in an 80 days period the variation in leaf positioning was within 0.05 cm, without requiring any recalibration. For a uniform 10×10 cm2 field, realized dynamically by a scanning 0.4×10 cm2 slit beam, a maximum variation in slit width of 0.01 cm was derived from ionization chamber measurements, both in time and for gantry rotation. For a clinical example, the dose in the reference point reproduced within 0.2% (1 SD) over a period of 100 days. Apart from regions with very large dose gradients, variations in the relative beam profiles measured with the EPID were generally less than 1% (1 SD). For different gantry angles the dose profiles also reproduced within 1%, showing that gravity has a negligible influence. No significant deviations between uninterrupted and interrupted treatments could be observed, indicating that the effects of acceleration and deceleration of the leaves are negligible and that a DMLC treatment can be finished correctly after a treatment interrupt. Our previous and present studies have demonstrated that the dosimetric accuracy and stability of intensity modulated beams, generated at the MM50 by means of dynamic multileaf collimation, are adequate for clinical use. Patient treatment using dynamic multileaf collimation has been started in our clinic. © 2000 American Association of Physicists in Medicine.