PURPOSE To determine the optimal parameters when using variance reduction techniques for the generation of treatment head phase space files for proton therapy dose calculations. To validate the doses generated from these phase spaces with respect to reference simulations. METHODS The treatment nozzles at the Francis H Burr Proton Therapy Center (setting 1) and at the University of California at San Francisco (setting 2) were modeled with the TOPAS Monte Carlo application. The geometry-based particle splitting technique (GPST) and the secondary production cut value (SPCV) were studied. For both settings, two particle-splitting planes were implemented. For setting 1 with a SPCV of 50 mm, the first plane was located upstream of the second scatterer, the other upstream of the aperture; the particles were split by a factor of 8 per plane. For setting 2 with a SPCV of 0.05 mm, the first plane was located upstream the water tank, the other upstream of the aperture and the split was by a factor of 16. Phase spaces in IAEA format were recorded and the normalized computational efficiency (NCE) was calculated. Full dose distributions in voxelized water phantoms for the reference and variance-reduced simulations were analyzed. In addition, dose calculation in a patient was simulated to compare the performance. RESULTS NCE values of 20 and 36 for the setting 1 and 2 were reached respectively. In all cases, percentage difference between dose profiles with and without variance reduction is within the statistical precision of 2%, 1 standard deviation. For the patient case, a NCE of 78 was reached while the 97.8% of the voxels passed the 3%/3mm gamma test. CONCLUSION By considering the GPST in combination with the SPCV, considerable gain in efficiency can be achieved without compromising the dosimetric accuracy. NIH/NCI under R01 CA 140735-01. CONACyT.