In the optically stimulated luminescence (OSL) dating of quartz, the presence of significant medium and slow components in the initial-OSL signal (first 0.8 s or so) can give rise to erroneous dose estimates. Thus it is desirable to develop a dose estimation method that is based on the OSL from the fast component alone. In this paper, the reduction of fast component OSL by high temperature infrared (IR) stimulation is investigated. It is shown that, in the temperature range from 120-190degreesC, it is possible to deplete the fast component preferentially using IR stimulation. The thermal assistance energy required for IR stimulation (1.49 eV; 830 nm) of the fast component is calculated to be 0.41 +/- 0.02 eV. However, the elevated temperature IRSL of quartz has a poor signal-to-noise ratio; hence a single-aliquot regenerative-dose (SAR) protocol based on a differential-OSL signal (the difference between the two short OSL signals separated by an elevated temperature IR stimulation) is proposed. The measurement conditions are optimised following examination of both the photoionisation cross-sections of the fast component for different stimulation temperatures, and possible thermally induced sensitivity changes arising from holding the samples at high temperatures for long periods (similar to 1500 s). The resulting equivalent dose values, and those from a conventional SAR protocol based on the initial-OSL signal, are compared with the expected doses for several samples. We observe that for samples with relatively strong medium and slow components, the SAR protocol based on isolation of the fast component (derived by IR depletion of the OSL) gives significantly more accurate estimates than the SAR protocol using net initial OSL signals. On the other hand, accurate dose estimates are obtained from samples dominated by the fast component using both protocols. Finally, the implications are discussed for dosimetry using quartz. (C) 2004 Elsevier Ltd. All rights reserved.