The article discusses several epistemological and methodological issues related to the analysis of discourse in general and of educational talk in particular. The theoretical framework provided by conversation analysis (CA) is applied and critically discussed in the analysis of an empirical example of educational talk. Several questions seem pertinent: Can we - as analysts - have direct access to talk "as it actually occurs", independent of any kind of theorizing and predefined categorization? What is the epistemological status of the conversation analytic categories? What are the limitations of applying turn-taking as an analytical category in the study of talk? To what extent can we presume the knowledgeability of the interlocutors as a premise in our analysis? On the background of my own attempts at applying CA in the analysis of educational discourse, I argue for a widening of the perspective from a narrow, empiricist focus on conversational turn-takings and sequential organization of talk, for example in the handling of issues like silences and absences in talk. On the other hand, I also warn against the pitfalls of historicist, abstract social theory; here exemplified with some texts from theorists applying abstract philosophical categories from dialectical and historical materialism like "the law of contradiction" as explanatory tools in the study of situated action. In the study of educational meaning making we should avoid empiricist as well as historicist approaches and explanations.