Quantitative content analysis is increasingly used to surpass surface level analyses in computer-supported collaborative learning (e.g., counting messages), but critical reflection on accepted practice has generally not been reported. A review of CSCL conference proceedings revealed a general vagueness in definitions of units of analysis. In general, arguments for choosing a unit were lacking and decisions made while developing the content analysis procedures were not made explicit. In this article, it will be illustrated that the currently accepted practices concerning the "unit of meaning" are not generally applicable to quantitative content analysis of electronic communication. Such analysis is affected by "unit boundary overlap" and contextual constraints having to do with the technology used. The analysis of e-mail communication required a different unit of analysis and segmentation procedure. This procedure proved to be reliable, and the subsequent coding of these units for quantitative analysis yielded satisfactory reliabilities. These findings have implications and recommendations for current content analysis practice in CSCL research.