Atmospheric dust is thought to have significant effects on the Earth's radiation budget, both directly and indirectly. However, modeling the global dust cycle is a difficult task due to the many complicated and nonlinear processes involved and the sparseness of observations for model evaluation. This paper describes a new off-line dust model, which can be forced by output from the Hadley Centre general circulation model (GCM) or by ERA analyses, and discusses an extensive set of quantitative and qualitative evaluations with DIRTMAP deposition data, University of Miami concentration data, and Total Ozone Mapping Spectrometer (TOMS) aerosol index. In general, the model agrees well with the observational data available, and the simulated annual mean global dust emissions of 2100 Mt yr-1 are in agreement with previous estimates. However, a suite of sensitivity studies shows that the model is highly sensitive to a wide range of ill-constrained parameters, particularly the threshold friction velocity and scavenging coefficient. Changes to these parameters result in changes in local dust concentration of up to a factor of 90. This raises issues about the reliability of model-estimated dust concentrations and the relative importance of different dust processes.