The so called classical artificial intelligence has always dealt with those high-level tasks which are still the privilege of humans, such as expert or commonsense reasoning, planning or natural language processing, whereas artificial neural networks appear to be much more suited to perceptual and low level tasks. Actually, weaknesses of symbolic systems correspond to the strong points of connectionist systems. So, from a rather technical point of view, it is very appealing to try to couple the two approaches in order to gather their best features. This paper points out the main types of hybrid systems, with a particular focus on those which have been applied to expertise modeling. Then it presents the SYNHESYS shell which has been developed at LIFIA, composed of a logic module (LM) and a connectionist module (CM), which has the following features: a symbolic/connectionist cooperation strategy to come to a decision, the possibility to learn continuously from examples, a CM-knowledge extraction procedure which allows the expert to validate the CM-knowledge and to correct or complete it before a new learning-extraction-validation cycle is undertaken.