This paper presents an overview of past results and ongoing work by the authors on voice source variation. The LF-model is used to quantify voice source parameters and ultimately as the basis for resynthesis. First of all, the nature and directionality of coarticulatory effects of voiced/voiceless segments are looked at across languages. Results show some striking cross-language differences, and some of these can be closely linked to the temporal coordination of laryngeal and supralaryngeal gestures. Voice source variations due to prosodic context (degree of stress) are also dealt with briefly. Concerning paralinguistic variation, ongoing work is presented on different voice qualities as produced by a single speaker. In conclusion, some implications for voice source rules in speech synthesis are mentioned.