Febrile illness is a leading cause of health-care seeking and hospital admissions in many settings. Estimates of the extent to which febrile illness could be attributed to different fever causing pathogens are sparse and when they do exist, are limited by studies with small sample sizes, small numbers of diagnostic tests for a small number of potential causes, no controls, and a lack of statistical approaches to use the data gathered and estimate attribution. The overall aim of this thesis is to further develop the application of Bayesian latent class models for investigating attribution of a syndrome to particular infections. Specifically, the aim is to provide estimates of the extents to which fever-related illness could be attributed to different fever-causing pathogens in four countries. The data used are diagnostic test results from fever cases and controls recruited in the Febrile Illness Evaluation in a Broad Range of Endemicities (FIEBRE) study. To meet these aims, methodological and applied work using Bayesian latent class models in two different but linked applications is carried out. The first estimates the accuracy of a pre-specified list of diagnostic tests in the absence of a perfect reference standard. This involves a simulation study to investigate the impact of the conditional independence assumption on estimates of diagnostic test sensitivity and specificity. This is an assumption made in simple latent class models. Then, diagnostic test accuracy meta-analysis are applied to estimate the accuracy of each diagnostic test of interest. The second application of latent class models uses the estimates of test accuracy from the meta-analyses as priors in a model with the observed multivariate imperfect binary diagnostic test data from cases and controls. The combined application of latent class models allows estimation of the fraction of fever cases attributed to different fever-causing pathogens from imperfect diagnostic tests.