For data with multicollinearity, penalized regression models i.e. Ridge regression (RR), least absolute selection and shrinkage operator (LASSO), elastic net (ENet) and adaptive LASSO (ALASSO) methods are popular alternatives to classical linear regression. The aim of this study was to comparatively examine performances of these models in order to predict the live body weight (BW) from various biometric body measurements and to select important variables in order to reduce model complexity. The data on body weight, withers height, body length, chest girth, paunch circumference, face length, length between ears, ear length, fat tail width and length were collected from 757 (247 male and 510 female) indigenous Harnai sheep of Pakistan. The present data were randomly partitioned into training and testing data sets and the hyperparameters of the penalized regression models were tuned using cross-validation. The performance of the studied models on both data sets were evaluated using the root mean squared error (RMSE), adjusted coefficient of determination () and Schwarz Bayesian information criterion (BIC) as goodness of fit criteria. The results revealed that all penalized regression methods provided accurate fit to the data. The ENet and ALASSO models were found to predict (on training data set) and (on testing data set) the BW with the highest accuracy for both male and female Harnai sheep.