The GrassGro decision support tool was designed to quantify sheep and pasture production in response to management and climate variability in temperate Australia, and has been tested in temperate but not low-rainfall Australian conditions. Data from field experiments and from on-farm monitoring was used to test GrassGro predictions of annual and perennial pasture production, and sheep production at four locations throughout the low-rainfall (275 to 375 mm annually) Victorian Mallee. Predictions of long-term pasture production were then made.Predictions of the herbage biomass of annual pastures closely matched observed data for both a sandy loam (1991-2002 data) and a whole paddock (combining sandy loam/loam and sand) (2001-2002 data) soil type, at several locations across the Victorian Mallee. Linear regression between observed and simulated data produced coefficients, significance and root mean square error of r2 = 0.81, P<0.001, 217 kg DM/ha, and r2 = 0.94, P<0.001, 72 kg DM/ha for sandy loam and whole paddock soil types, respectively. A series of simulations for individual years between 1970-2002 quantified the large impact of climate variability and demonstrated that seedbank and location, but not soil fertility, have a large influence on annual pasture production. However, GrassGro underestimated the production of the perennial pasture, lucerne (r2 = 0.2). GrassGro was also unable to adequately predict sheep production because it failed to take into account the sparse, clumpy structure of the low biomass pastures typical of this region. Methods to improve GrassGro were identified and include: a) the need to adjust sheep intake from low biomass, sparse pastures, b) the ability to predict summer growing and autumn growing plant species, c) the ability to graze crop stubbles, and d) refinements to the coefficients of equations used to model lucerne growth.