In an era of rapid advancement in analytical techniques the amount of data that needs to be properly processed increases. According to the European Network of Forensic Science Institutes, a proper interpretation of data for forensic applications should be embedded in a likelihood ratio (LR) framework. The method clearly reflects the role of the forensic expert in the process of evidence evaluation. The concept involves analysis of the evidence data in the context of two adversative hypotheses, e.g. the sample recovered from the suspect's clothing and the sample collected from the crime scene may have come from the same object (H1), or not (H2). The LR model evaluates the similarity between the samples, the frequency of observing their data and typical variability within and between such samples in the population to indicate which of the hypotheses is more likely. The chapter focuses on hybrid LR models, which were developed to bypass the infeasibility of training LR models for datasets with more variables than samples. They are constructed for a limited number of variables derived from chemometric techniques that effectively reduce data dimensionality, enhance the differences between samples in the training set and reduce the variance within them for improving the performance of LR models.