There exists a number of semi-empirical relationships assessing void ratio in the literature. The predictive performances of widely used void ratio relationships are assessed within a probabilistic framework with the use of compiled database. The database, which consists of 636 non-plastic silts, sands, and gravels is compiled from the existing literature having minimum, maximum void ratio, mean grain size, coefficient of uniformity, particle roundness and sphericity, fines content parameters. Among predictive models, which use D-50 as an input parameter, Chen and Kulhawy model produced the least error (mu(error)=-0.006, sigma(error)=0.125) and is concluded to be more accurate compared to Cubrinovski and Ishihara model (mu(error)=-0.055, sigma(error)=0.140). Similarly, Zheng and Hycriw, which uses C-u as an input parameter produced more accurate predictions (mu(error)=-0.034, sigma(error)=0.069), when compared to Chen and Kulhawy model (mu(error)=-0.062, sigma(error)=0.137).