We present an overview of the paradigm of learning-based testing (LBT) for software systems. LBT is a fully automated method for specification-based black-box testing using computational learning principles. It applies the principle of tests as queries, where queries are either generated by a learning algorithm or by a model checker through use of a formal requirements specification. LBT can be applied to automate black-box testing of a variety of different software architectures including procedural and reactive systems. We describe some different testing platforms which have been designed using this paradigm and some representative evaluation results. We also compare LBT with related testing methods.