Technical Abstract: Fiber length is a very important factor in yarn production. A study was carried out to investigate the effects of various cotton fiber length parameters (conventional and non-conventional) and their combinations on yarn properties. Linear regression models with different number of fiber length parameters and their combinations were developed to predict different properties of yarns (ring, OE). Our results indicate that, for the models with only length parameters, four-parameter models are more proper models considering their R2s and Mallow’s Cp values. If other fiber properties (strength, Mic, etc.) are included, the R2 can be further increased. Short fiber parameters and length CV are important for improving prediction accuracy of yarn properties. Lower Half Mean Length (LHML) predicts yarn properties similar to short fiber content (SFC). Best prediction models for different yarns (ring, OE, etc.) and different yarn properties (strength, irregularity, etc.) include different length parameters. Not all yarn properties can be predicted well by linear regression models with length parameters. For yarn properties such as neps, ends-down, elongation, and CV (coefficient of variation) of strength, all models had low R2s.