Wheat (Triticum aestivum L.) is one of the most important crops for food production in the world. The importance of wheat emphasizes the fact that wheat products are the most important source of dietary proteins and energy supply for humans. Therefore, achieving suitable wheat quality is of great importance. Wheat quality is determined by a large number of traits and under the strong environmental influence. One of the most important factors affecting dough rheology is gluten content and its strength. Gluten is the most abundant wheat protein, and by its structure, gluten is a complex network of monomeric gliadins and polymeric subunits of glutenin. Among gluten components, high molecular weight glutenin subunits have the greatest impact on dough quality. Different instruments can be used to perform rheological tests, which are necessary to assess wheat baking quality more accurately. Mixograph is a dough mixer that creates a dough rheological profile, providing general information about dough mixing, its behaviour during development, and the strength of the dough. Due to the small amount of flour required, mixograph has shown to be highly suitable for use in wheat breeding, particularly in early generations when availability of grain and flour is still limited. Since the majority of quality traits have complex inheritance patterns, breeding for improved baking quality is one of the most demanding objectives in wheat breeding. Taking that into account together with often costly and time-consuming phenotyping, predictability of wheat baking quality may be very challenging. The ability to develop a genotype that exhibits both improved performance and high stability of the quality traits is critical to the success of wheat quality improvement. One of the major challenges in plant breeding, in this context, is the occurrence of genotype-by-environment interaction, since its presence makes selecting widely adapted genotypes difficult. The AMMI model is one of the most commonly used methods for the analysis of genotype-by-environment interaction. However, the extensive development of high-throughput genotyping in the last decade has enabled reliable and rapid predictions of breeding values based only on marker information. Genomic selection is one of the recently developed methods that enables the prediction of breeding values of individuals by simultaneously incorporating all available marker information into a model. Genomic selection aims to capture total additive genetic variance based on the sum of the effects of a large number of genetic markers, encompassing all QTLs that contribute to trait variability. In genomic selection, genotypic and phenotypic information of the training population is used to train a model and estimate the marker effects. Obtained data is then applied to the breeding (validation) population of non-phenotyped candidates to estimate their genomic-estimated breeding values (GEBV). The effectiveness of genomic selection is determined by the obtained prediction accuracy, which is affected by a variety of molecular, genetic, and phenotypic factors, as well as the parameters of the selected statistical model. The correct adjustment of factors that can affect prediction accuracy, such as population structure, size of training population, the relatedness of training and validation population, marker density, etc., is the first step toward successful implementation of genomic selection in practical breeding programs. Different prediction models have been developed to solve the problem of high-dimensional datasets occurring in genomic selection. These models differ primarily in their assumptions about the distribution and variance of marker effects, i.e., how marker effects contribute to the trait. Given the often challenging phenotyping for wheat quality traits, and especially for baking quality traits, the use of classical breeding methods can be costly and time-consuming. Determining the optimal model and parameters of genomic selection would enable the use of molecular markers in the pre-selection process for grain quality traits and the optimization of classical wheat breeding methods. This research aimed to assess the impact of genotype-by-environment interaction and optimize genomic selection for grain quality traits using biparental wheat populations, in order to reduce the potential costs of x genotyping and phenotyping in the breeding process and suggest optimal strategies based on genomic selection for more efficient development of new lines. Two biparental populations of winter wheat were used in this study. The BK population was derived from the Bezostaya-1 × Klara cross and the MG population from the Monika × Golubica cross. In the BK combination the parental genotypes differed in all high molecular weight glutenin subunits, while the parental genotypes used in the MG combination did not differ in any of the high molecular weight glutenin subunits. The BK and MG populations consisted of 145 and 175 genotypes, respectively, including parental genotypes. Field trials were conducted for three consecutive years (2009 – 2011) at two locations in Croatia (Osijek and Slavonski Brod), i.e., in six different environments. In each environment the field trial was set up according to a row-column design. Genotyping of both populations was done using Diversity Arrays Technology. After marker filtering the final dataset used for genomic selection contained 1087 and 2231 SNP markers for BK and MG population, respectively. Grain protein content, wet gluten content, test weight, and dough rheology were assessed in each environment. Dough rheology was investigated using a mixograph and four variables were selected for further statistical analysis (MPT, MTW, MTI, and MPH). Genotype-by-environment patterns for the quality and mixograph traits were studied using the AMMI model. The dissection of genotype-by-environment patterns was visualized by a modified version of the AMMI2 biplot, which adds the main effects to the standard AMMI2 biplot using a colour scale. In the first phase of the genomic selection analysis the need for optimization of the training population based on phenotypic variance was assessed using both biparental populations. Additionally, the influence of the training population size and marker density on the prediction accuracy was investigated. For that purpose, three different sizes of training population were used for both BK and MG populations, and two different marker densities for the MG population. The first phase was conducted using only the RR-BLUP model. In the second phase of the genomic selection, the performance of seven different genomic selection models was compared with the performance of the RR-BLUP model. Models included were elastic net, four Bayesian models (BayesA, BayesB, BayesC, and BayesLASSO), random forest, and reproducing kernel Hilbert spaces. This part of the analysis was performed only within the MG population. Results revealed some positive as well as negative transgressive segregants in both populations for all quality traits although being generally more prevalent in the BK population. This may suggest the dispersion of the alleles with positive (increasor) and negative (decreasor) effects between parental genotypes in both crosses. The environment was the dominant source of variation for grain protein content, wet gluten content, and test weight, accounting for approximately 40% to 85% of the total variation. The pattern was less consistent for mixograph traits for which the dominant source of variation was trait- and population-dependent. Overall, genotype-by-environment interaction was shown to play a more important role for mixograph traits compared to other quality traits. Inspection of the AMMI2 biplot revealed some broadly adapted RILs, among which MG124 is the most interesting, being the prevalent “winner” for grain protein content and wet gluten content, but also the “winner” for non-correlated trait test weight in environment SB10. The broad-sense heritability across environments was high for all traits, except for MPT in the BK population, the heritability of which was 0.45. Although repeatability varied considerably among environments, it was high for most of the trait-environment combinations, with a value above 0.7. These results suggest that heritability itself should not represent a limitation in achieving good prediction accuracy. The results of genomic selection analysis showed that the size of the training population plays an important role in achieving higher prediction accuracies, while marker density does not represent a major limitation. Additionally, the results of the present study did not support the optimization of the training population based on phenotypic variance as a tool to increase prediction accuracy. The performance of eight prediction models was compared and among them, elastic net showed xi the lowest prediction accuracy for all traits. Bayesian models provided slightly higher prediction accuracy than the RR-BLUP model. However, this may be considered negligible considering the time required to perform an analysis. Although RR-BLUP was not the best performing model in all cases, no advantage of using any other model studied in this research was observed. Furthermore, strong differences among environments in terms of the prediction accuracy were observed. For example, the prediction accuracy for TW within the MG population was moderate in one environment, while being low in all other environments. Comparing these results to the results of a genotype-by-environment analysis it is noticeable that environments that are characterized by unusually high or low values for prediction accuracy compared to the rest of the environments tend to be those that produce the greatest genotype-by-environment interaction. This suggests that less predictive environments should be excluded from the dataset used to train the prediction model in order to achieve higher prediction accuracies. The prediction accuracies obtained in this study support implementation of genomic selection in wheat breeding for end-use quality, including some mixograph traits.