Performance evaluation of a software system is often conducted under uncertainty, mainly due to unpredictable factors such as fluctuating workloads and availability of software and hardware resources. Uncertainty is particularly critical in performance engineering when it relates to the values of important parameters such as workload, operational profile, and resource demand, because such parameters inevitably affect the overall system performance. So far, this research area has focused on monitoring the performance characteristics of software systems while considering variable configuration options, however the problem of incorporating uncertainty as a first-class concept in the software development process to identify performance issues is still challenging. The PLUS (Performance Learning for Uncertainty of Software) approach aims at addressing these limitations by investigating the specification of a new class of performance models capturing how the different uncertainties underlying a software system affect its performance characteristics. The main goal of PLUS is to answer a fundamental question in the software performance engineering domain: How to model the variable configuration options (i.e., software and hardware resources) and their intrinsic uncertainties (e.g., resource demand, processor speed) to represent the performance characteristics of software systems? This way, software engineers are exposed to a quantitative evaluation of their systems that supports them in the task of identifying, from a performance perspective, the most critical configurations along with their uncertainties leading to generate performance issues.