Sixteen children in five different classrooms in the same, small, rural-suburban school in New Hampshire were observed for 2 years to document what primary children did when they wrote. Eight of the children were observed from age 6 through 7, and eight from 8 through 9. Three researchers were on-site in classrooms 4 days out of 5 for the 2-year project. Data were gathered through child and teacher interviews, direct observation of children through specific protocols, and video recordings of children while composing, conversing with other children, and in conference, as well as through all of the children's written products. Data from observations of child behavior during the writing process together with data from observations of teacher practices led to seven hypotheses: (1) behaviors of writers are ideosyncratic and highly variable; (2) clusters of behaviors should be observed before making decisions about writers; (3) scope and sequence curricula have little relevance in helping writers develop; (4) the scaffolding-conference approach is the best response to the variable writer; (5) let students write daily, sustain selections longer, and write at predictable times; (6) let children choose most of their topics because it assists them with voice, heightens semantic domain, skill of narrowing topic, and basic decision making; and (7) skills are best taught within the context of the child's own writing. (Appendixes include copies of research articles, selected chapters from a book, articles related to the study, and data examples.) (HOD)