This thesis presents a mathematical modeling approach to estimate the contribution of four animal reservoirs of the food chain to the occurrence of salmonellosis cases in humans in the European Union. In addition, an alternative and more explorative approach based on expert elicitation is attempted in order to extrapolate results to countries with less data availability, as a first step to perform source attribution of Salmonella in a more global perspective. <br/>Cases of foodborne salmonellosis in humans were attributed to travel, outbreaks and four animal reservoirs, namely pigs, broilers, turkeys and laying hens, using a Bayesian model based on microbial subtyping in 24 countries of the European Union. The chosen approach is recognized as data intensive, requiring numbers for Salmonella occurrence in food-producing animals, reported human cases, information on possibility of infection abroad (from here on referred to as “travel information”), human cases originating from outbreaks with and without a confirmed source and amounts of the meat or eggs of each animal reservoir originating from each country and available for consumption in each country. Thus, special data management, analysis and validation was required to produce a dataset containing standardized information for all countries (Manuscript I). <br/>Data on reported human cases were provided by the European Centre for Disease Prevention and Control (ECDC) through the European Food Safety Authority (EFSA). Salmonella prevalences in animals were obtained from the EU-wide baseline studies (BS) conducted by EFSA and complemented where necessary with information found in the European Union Summary Report (EUSR), as published by EFSA. Information on outbreaks was also provided by EFSA. The amount of food available for consumption was calculated based on trade data obtained from the European Statistical Office (EUROSTAT) and complemented with information from the Association of Poultry Processors and Poultry Trade in the European Union Countries (AVEC). Common limitations included non-participation in all BS, non-reporting of outbreaks or travel information, non-reporting of serovar-specific information, non-reporting of case-based data and non-availability of trade data on EUROSTAT. In order to standardize the information available, cases without travel information were assumed to be domestic; cases without specific serovar information were redistributed according to serovar proportions observed in the same dataset or other reference documents; missing trade information was estimated based on previous years, and non-participation in a BS was supplied, where possible, with data from the EUSR. When the lack of original data was considered too extreme to the point of compromising the attribution results, countries were excluded. The resulting dataset comprised Austria, Belgium, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, the Netherlands, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden and the United Kingdom. Three countries were included in the initial analysis, but were excluded from the final dataset. Those were: Bulgaria, which presented 100% of human cases without serovar detailing; Romania, which only participated in one BS and had not enough surrogate data to be retrieved from the EUSR, besides reporting a large parcel of cases without serovar information; and Norway, which is not part of the EU and does not report to EUROSTAT (Manuscript I). <br/>A Bayesian modeling approach which compares the occurrence of serovars in humans with the occurrence of the same serovars in animals of the food-chain was used to estimate the contribution of each of these reservoirs, travel and outbreaks to the number of human cases of salmonellosis in the 24 countries present in the dataset previously described (Manuscript II). Laying hens (i.e. eggs) were estimated to be the most important source of human salmonellosis at EU level, with 48.1% (95% Credibility Interval (CI) 47.5 – 48.8%) of cases, followed by pigs (29.6%, 95% CI 28.9-30.3%). Turkeys and broilers were estimated to be less important sources of Salmonella, contributing with 4.4% (95% CI 4.2-4.7%) and 3.7% (95% CI 3.4-4.0%), respectively. A total of 10.2% of all salmonellosis cases were reported as being travel-related, and 3.9% of cases were reported as being part of outbreaks with unknown source. S. Enteritidis was the most important serovar in the study, being responsible for 95.9% of cases attributed to laying hens, 56.9% of cases attributed to broilers, 30.4% of turkeys and 28.3% of cases attributed to pigs, for which the main serovar was S. Typhimurium (63.1% of cases attributed to this source). Country-specific results show laying hens as the most important source of salmonellosis in 13 countries (Austria, Czech Republic, Estonia, Germany, Greece, Hungary, Latvia, Lithuania, Luxembourg, Slovenia, Slovakia, Spain and the United Kingdom), whereas pigs were the larger animal contributor in eight (Belgium, Cyprus, Finland, France, Ireland, Italy, Poland and Sweden). In Finland and Sweden the majority of Salmonella infections were estimated to be travel-related. Travel was also an important source in Ireland, the UK and Denmark, although to a lower extent. In the Netherlands, the proportion of disease attributed to layers and pigs were similar. In Denmark, the most important food-animal source was estimated to be turkeys, and broilers were the major source in Portugal. (Manuscript II). <br/>Danish strategies for risk management of Salmonella in the farm-to-fork continuum include the routine application of a source attribution model to estimate the contribution of the major animal-food sources to human infections by Salmonella in Denmark. This model concept formed the basis for the model described in Manuscript II. As part of the validation process of the EU model, results for Denmark in the EU model were compared with the ones obtained using the Danish model in the same period (Manuscript III).The Danish model points to pork as the main animal source of human salmonellosis in the period (9.3% of cases), followed closely by table eggs (7.5% of cases) and broilers (4.7% of cases), while the EU model attributed 15.6% to pigs, 15.1% to turkeys, 10.5% to eggs and 2.8% to broilers. Travel-related cases constitute 30.6% in the Danish model and only 18.2% in the EU model. Cases that could not be attributed to any source corresponded to 16.7% in the Danish model and 14.1% in the European model. Discrepancies in numbers are explained by differences in model structure and basic assumptions: a) cases with no travel information in the Danish model are redistributed according to proportions observed in cases with full information; in the EU model, as some countries did not provide any information regarding travel prior to sickness, it had to be assumed that no information means no travel; b) the Danish model uses data subtyped to phage-type level, while the EU model only uses serovar level, as phage-type data in humans and animals was not sufficiently available; this allows the more specific allocation of cases to the right sources; c) the larger number of sources in the Danish model allows more options for specific allocation of cases, presumably resulting in a more correct distribution of cases among sources; d) the Danish model uses official data on amount of domestic and imported food items available for consumption in the country, but does not as opposed to the EU model take into account the amount imported from each country specifically, which results in an underestimation of the contribution from high prevalence countries as compared to the EU model (Manuscript III). All facts considered, the two models rank three out of the four sources in a similar order and, while the EU model is considered useful for countries which cannot readily attain the level of detailing found in Denmark for monitoring and surveillance data, Denmark would benefit more from applying country-specific data than to adopt the results of the EU model. <br/>The last chapter presents an alternative approach to obtain results for the Czech Republic, Bulgaria, Norway and Romania, the last three of which were excluded from the EU-model due to insufficient data. Using clustering techniques, 28 countries were grouped according to variables used to characterize them as to social and economic status, animal production characteristics and food consumption patterns. Where available, variables reflecting the occurrence of Salmonella enterica in humans and animals were also used. The results of the analyses were delivered to a panel of experts composed by foodborne disease epidemiologists and risk modelers, which were asked to provide attribution estimates for the aforementioned countries, based on their similarity to countries for which results were previously obtained. Experts were also asked to evaluate the method concerning its utility and applicability of results. Individual estimates were evaluated based on comparison with the Czech results, for which results based on the microbial subtyping model were available, but also in relation to uniformity of guesses and uncertainty intervals among different estimates from the same expert and among all experts in the panel. This evaluation resulted in five out of the seven respondents being maintained in the panel. Although the Czech Republic values obtained did not match the ones observed in the EU study, the order of importance of the animal sources was in agreement between the two studies and there was also a consensus in the panel concerning that order. It is, therefore, believed that with some adjustments, this method may be useful for prioritizing targeted actions for Salmonella control in countries without sufficient data for a traditional approach. Further on, this method may be used to identify “surrogate countries” from where animal prevalence data can be “borrowed” and applied in the traditional microbial subtyping approach in the aforementioned Member States. <br/>This PhD project has provided results for a European “source of infection account” for Salmonella, and has at the same time been evaluating the approaches attempted, raising questions and proposing solutions on how to deal with the lack of good-quality data for such studies. The project has also achieved results that may lay the groundwork for future attempts to develop Salmonella source attribution estimates in a more global perspective.