An improvement of the software industry's software development effort estimation processes may benefit from a better understanding of the mental, partly unconscious, processes involved in estimating effort. This paper proposes and tests a theory potentially explaining essential parts of typical judgment-based effort estimation processes (“expert estimation”). The theory is based on findings from the human judgment research literature and proposes that judgment-based effort estimation is based on: i) an early categorization of the project to be estimated, ii) a resistance towards a change of the chosen category, and, iii) a “regression” of the effort estimate towards a reference value of the chosen category, where the amount of regression depends on the level of uncertainty of the project work. Implications of the theory are tested with results from three software effort estimation experiments. All examined studies confirmed the theory. There is, however, a strong need for more work, independent evidence and clearer description of scope and concepts part of the theory. Finally, we outline a study planned for further testing of essential parts of the theory.