Fire managers around the world commonly use visual assessment of forest fuels to aid prediction of fire behaviour and plan for hazard reduction burning. In Australia, fuel hazard assessment guides also allow conversion of visual assessments to indicative fuel loads, which is essential for some rate of spread models and calculation of fireline intensity or emissions. The strength of correlation between fuel hazard and destructively sampled (directly measured) fuel load was tested using a comprehensive dataset of >500 points from across a range of eucalypt forests in Australia. Overall, there was poor correlation between the assigned fuel hazard rating and measured biomass for surface, near-surface and elevated fuel components, with a clear tendency for these systems to under-predict fuel load at low hazard ratings, and over-predict it at high hazard ratings. Visual assessment of surface fuels was not statistically different from a random allocation of hazard level. The considerable overlap in fuel load between hazard ratings at higher ranges suggests the need to reduce the number of hazard classes to provide clearer differentiation of fuel hazard. To accurately assess forest fuel condition, improvements in fuel hazard descriptions and calibration of visual assessment with destructively measured fuels is essential.