A parallel-plate viscometer has been designed for use in an internally heated pressure vessel (IHPV) at pressures up to 350 MPa and at temperatures up to 900 degrees C. The viscosity of a melt is determined by measuring the rate of deformation of a cylindrical sample as a function of an applied, constant stress at a fixed temperature. The viscometer consists of a small furnace with two independent heating resistors, a moveable load by which the stress is applied to the sample, and a pressure-resistant transducer (LVDT) that measures the deformation of the sample. The accessible viscosity range covers three orders of magnitude from 10 (super 8.5) Pa.s to 10 (super 11.5) Pa.s. Calibration measurements on the standard melt DGG1 at 0.1 MPa demonstrated the precision of the viscometer to be within + or -0.08 log units. Subsequent measurements at elevated pressure on DGG1-melt, Di 100 -melt (Di = CaMgSi 2 O 6 ), and Ab 55 Di 45 -melt (Ab = NaAlSi 3 O 8 , composition in weight percent) showed a pronounced increase of viscosity with pressure. Comparison with literature data on the pressure dependence of the transformation temperature of Di 100 -melt (Rosenhauer et al. 1979) confirmed the reliability of these findings. The dependence on pressure becomes smaller with increasing temperature for these depolymerized melts; e.g., in the case of Di 100 -melt (NBO/T = 2) from deta /dP = +0.23 log units/100 MPa at 751 degrees C to deta /dP = +0.18 log units/100 MPa at 770 degrees C. In contrast to the depolymerized melts, a polymerized melt of haplotonalitic composition (NBO/T = 0) shows a decrease by -0.12 log units/100 MPa in the pressure range 50-350 MPa at 889 degrees C. Possible application of the new viscometer to study rheological properties of volatile-bearing melts was tested successfully with a hydrous haplotonalitic melt. Addition of 3.80 wt% of water to the anhydrous melt strongly shifts the viscosity-temperature relationship toward lower temperatures; e.g., at a viscosity of 10 (super 10.5) Pa.s from 883 to 515 degrees C. The measured viscosities did not drift during the run, indicating that water loss is negligible within the time scale of the experiments, as confirmed by IR-microspectroscopic analysis.