Many real-world optimisation problems often feature multiple conflicting objectives that must be optimised simultaneously. For a subset of these problems, there are no analytical forms or numerical simulation models for the objective functions. Yet, there exist datasets relating the input parameters to the associated objective values. In such cases, the only realistic option is to deploy surrogates, also known as metamodels or emulators of the objective functions, alongside a multi-objective evolutionary algorithm (MOEA) to locate a good approximation of the Pareto front. Since there is no opportunity to gather new data, it is important that we quantify the uncertainties associated with surrogate predictions and utilise them during optimisation and communicating solutions to decisions-makers. In this chapter, we summarise the challenges of using surrogates with MOEAs to solve data-driven optimisation problems. We mainly focus on the probability of selection criterion in decomposition-based MOEAs and show its potential on several benchmark problems with different settings.