Yield is a key trait in pepper and is often measured simultaneously with other traits over several environments. The understanding and genetic improvement of yield may benefit from the joint analysis of yield with its related traits simultaneously. Linear mixed models have emerged as a flexible approach that correctly model underlying variance-covariance structures among the traits and between environments simultaneously. ln this study, we applied four different QTL approaches based on linear mixed model on five yield related pepper traits measured across four environments. We evaluated the performance of the approaches in terms of the number of QTLs detected for each trait and their explained variance. The QTL models are a single-trait single-environment approach (STSE), a multi-trait approach (MT), a multi-environment approach (ME), and a multi-trait multienvironment approach (MTME). We further compared prediction accuracies between STSE and MT models. The predictions were subjected to a five-fold cross validation. Our results showed that multi-trait and/or multi-environment QTL analyses are more powerful and effective to map pleiotropic QTL and QTL by environment interactions th an performing STSE analysis. The multivariate models further showed improvement over STSE in terms of both number of QTLs and the explained variance. MTME clearly outperformed ail the other methods. With MTME, nine QTLs explaining 51% of genetic variation were identified for yield in the autumn trial in Spain as against three, three and six QTLs explaining 37%,29% and 43% from STSE, ME and MT analyses, respectively. The MT model for yield in SP2 had prediction accuracy of 0.53, against 0.42 from the STSE model. These results confirmed that multivariate analyses of traits have better capabilities to unravel complex traits than single trait approach. Our result showed that trait's prediction accuracy depends not only on prediction model of choice and traits genetic architecture but also on the environmen