In recent decades, animal breeding science has focused primarily on improving performance traits, but animal breeding is continuously evolving. Thus, the current focus of breeding goals is generally oriented towards social interests for more animal welfare and sustainability. In addition to traits that are of direct monetary importance for practice, such as milk yield, milk ingredients (protein, fat), fattening performance, laying performance, and so on, traits that only have an indirect monetary evaluation or currently have no economically remunerated value at all have increasingly played a role in recent years. These include traits such as meat quality, animal behaviour, robustness, resistance to diseases or auxiliary traits such as somatic-cell-count, milk ketone bodies (ß-hydroxybutyrate, acetone) or milk fatty acids as indicators of (metabolic) diseases. By identifying and establishing new traits, animal breeding science tries to contribute to sustainable agriculture. In practical breeding work, however, the estimated breeding values and their reliability/accuracy are decisive for successful breeding. The earlier the farmers have access to breeding values with appropriate reliabilities/accuracies, the more efficiently they can make selection decisions. Within large populations, genomic selection based on large training samples is now well established, providing breeding values, although initially with moderate reliabilities/accuracies, at a very early stage. Endangered breeds or small populations face a problem here, since the population structure usually does not provide the necessary size for the training sample. Furthermore, the financial effort for genotyping is usually quite high. However, especially these small and partly endangered breeds are the living genetic back up for animal breeding sciences, in order to be able to cross new traits into highly specialized breeds in the future and thus meet market requirements. The aim of the present work is to optimize the classical pedigree-based breeding value estimation for small-structured populations, especially endangered breeds, in such a way that higher reliabilities/accuracies are generated for breeding practice and a ranking of genotypes within production systems (= farm types) is enabled. For this purpose, 30 cattle farms with “Deutsches Schwarzbuntes Niederungsrind” (DSN) or “Holstein Friesian” (HF) and 45 pig farms with “Bunte Bentheimer” pigs (BB) were first characterized based on socio-ecological criteria. Based on the assumption that phenotypic performance of animals is influenced by environmental and management factors of similar farms at a similar level, different cluster methods (agglomerative hierarchical clustering, partitioning around medoids, fuzzy clustering and clustering of variables combined with agglomerative hierarchical clustering) are used to identify similarities. The goal is to group farms with similar characteristics within farm types. In both the cattle and pig populations, clustering of variables combined with agglomerative hierarchical clustering (CoVAHC) based on silhouette width (= evaluation criterion) proves to be the best method for grouping farms into farm types. Four farm types can be identified as optimal in the cattle dataset and three farm types in the pig dataset. Based on the recorded characteristics, the cattle farm types can be differentiated as “medium DSN farms with a focus on milk production”, “small DSN farms with low intensity”, “large intensive DSN farms” and “specialized HF farms”. The focus of characterization of the pig farms is mainly on their breeding activities as well as their marketing potential. Accordingly, the farm types can be described as farm types with low, medium and high breeding activity or with low, medium and high marketing potential. The recorded production and fertility traits. Health indicators, and meat and carcass quality traits, respectively, show predominantly highly significant differences between the individual farm types of the respective species, which underlines that the grouping of farms into farm types is reasonable. Within the cattle population, the use of a common data set (DSN and HF), taking into account a breed effect in the modelling, can generate 1,5 to 3,0 times higher reliabilities as well as higher goodness of fit of the modelling, compared to separate breed-specific calculations. The different definitions of the contemporary groups (farm, farm type, combination or nesting variants of both, as well as combination with test day or test month) in the cattle data set show a clear advantage of the formation of contemporary groups by means of farm type as a single effect in combination with the test day or the test month, resulting in the highest overall reliabilities as well as goodness of fit. The difference between the two combination effects test day or test month leads only to minor differences in the evaluation criteria. The calculation of the genetic parameters (variance components, heritabilities as well as breeding values and their reliabilities/accuracies) basically show a superiority when using farm types instead of the farm effect in the modelling. Thus, increases in heritability of up to 16 % (protein quantity; test day) and 17 % (protein quantity; test month) in the cattle population and up to 11 % (conductivity) for the classical traits and up to 17 % for the meat quality traits recorded in vivo in pigs can be observed compared to the model with farm effect. The reliabilities/accuracies of the estimated breeding values can also be increased considerably in most cases. Thus, for the production traits milk and protein yield, increases in reliabilities between 3,9 % (daily milk yield in the total population, combining farm type with test day) and 9,8 % (daily protein yield in the bull population taking into account the test month) were achieved. In the pig population, depending on the trait, increases in accuracies of noticeably more than 10 % up to 39 % (back fat percentage recorded in vivo within the boar population) can be reported. Although heritabilities and reliabilities/accuracies for the functional traits can also be partially improved by using the farm types, they are at a considerably lower level. Here, the small size of the dataset, the low assessment frequency and the lack of data structure/distribution for the traits body condition score, cleanliness udder, lameness, methane emission (cattle) and behaviour, fundament and exterior (pigs) have a negative effect on the estimated parameters. Nevertheless, increases in reliability of up to 2,4 % (linear modelling, cleanliness udder, test month) and increases in accuracy of up to 22,3 % (linear modelling, fundament, boar population) can be achieved. However, the farm type effect also leads to a decrease of genetic parameters for some traits, e.g. within the cattle data set for the trait body condition score in the amount of -0,6 % (linear modelling, test month) or in the pig data set for the trait exterior of -9,5 % (logistic modelling, boar population). Within the cattle population, it can be seen that especially when only few daughter information (< 15 progeny performances) is available, the farm type effect allows significant increases in reliabilities (milk quantity: up to 9,7 % (test day) and up to 11,2 % (test month)). In contrast, the consideration of test month compared to test day causes only slight increases in breeding value reliabilities. In the preliminary reflections of this work it was argued that the use of farm types in the genetic-statistical modelling leads to a reduction of residual variance, but there was rather an increase of additive-genetic variance. Based on the analysis of the population structure, it can be shown that the use of farm types basically leads to an increase in the number of bulls/boars used within farm types compared to single farm analysis. Also, the number of bulls/boars with compared to bulls/boars without progeny information is increased. Furthermore, the inbreeding and relationship coefficients being averaged over the farm types are lower compared to those in individual farms. Consequently, the farm type effect causes a more even distribution of sires, which ultimately leads to a higher estimation accuracy. The identified genotype-environment interactions in both the cattle and pig populations clearly show that not every animal is optimally adapted to the respective farm types. This is also confirmed in the ranking differences of the top sires. Here, based on the level of the determined breeding values, there are sometimes considerable shifts in the ranking of the sires. In summary, modeling using farm type effect is superior to modeling using the farm effect in both the small populations (DSN and BB) and large populations (HF) in almost all characteristics studied.