Laughter is a non-verbal phenomenon, widely used in human interaction, which has been shown to differ from speech along various acoustic-prosodic dimensions. Previous work has also revealed that the production of laughter is subject to a high degree of variation, with speakers normally having several types of laughter in their repertoire. Despite this, relatively little is known about how different types of laughter are marked prosodically and whether prosodic features may be used to discriminate between laughter types. We investigated here two types of laughter events produced in spontaneous interaction, with respect to five prosodic characteristics: duration, pitch, intensity, rhythm and voice quality. Our results showed that each of these characteristics, except for rhythm, differ between laughter types. We then employed prosodic features in a machine learning system trained to discriminate between three classes: speech and the two types of laughter. The proposed system obtained a similar speech/laughter classification performance to that of a system that considers only two classes, speech and laughter, while also having the advantage that a finer distinction, i.e., between laughter types, may be achieved.