Model-driven Engineering (MDE) approaches are often recognized as a solution to palliate the complexity of software maintainability tasks. However, there is no empirical evidence of their benefits and limitations with respect to code-based maintainability practices. To fill this gap, this paper illustrates the results of an empirical study, involving 44 subjects, in which we compared an MDE methodology, WebML, and a code-based methodology, based on PHP, with respect to the performance and satisfaction of junior software developers while executing analysability, corrective and perfective maintainability tasks on Web applications. Results show that the involved subjects performed better with WebML than with PHP, although they showed a slight preference towards tackling maintainability tasks directly on the source code. Our study also aims at providing a replicable laboratory package that can be used to assess the maintainability of different development methods.