The physical characteristics of a fiber determine its processing behavior, production efficiency and finally yarn and fabric quality. Therefore, predicting the quality characteristics of yarns, such as the tensile properties from the raw material properties, was the main purpose of many studies in the last century. In addition to raw material processing conditions, preparation stages, machine parameters and the spinning method also have considerable effects on the yarn properties. Generally two approaches were used in these studies for predicting yarn quality from fiber and yarn characteristics:• an empirical and statistical approach; and • a theoretical or analytical approach. 1The empirical and statistical approach to establishing a relationship between fiber and yarn quality characteristics has been the most popular method during the second half of the twentieth century. Fast and accurate measurement of fiber properties by means of high volume instruments (HVI) and more powerful computers are the two main reasons for this tendency. With this method, important fiber and yarn properties can be measured for a range of samples and by using these results empirical relationships have been established by means of statistical analysis. One of the Abstract The main aim of the present study was to predict the most important yarn quality characteristics derived from cotton fiber properties that were measured by means of an HVI system. With this aim 15 different cotton blends were selected from different spinning mills in Turkey. The cotton fibers were processed in the short staple spinning line at Ege University Textile and Apparel Research-Application Center and were spun into ring yarns (20s, 25s, 30s and 35s). Each count was spun at three different coefficients of twist (α e 3.8, α e 4.2 and α e 4.6). Linear multiple regression methods were used for the estimation of yarn quality characteristics. Yarn count, twist and roving properties all had considerable effects on the yarn properties and therefore these parameters were also selected as predictors. After the goodness of fit statistics very large R 2 (coefficient of multiple determination) and adjusted R 2 values were observed. Furthermore, analysis of variance tables showed that our equations were significant at the α = 0.01 significance level.