ユーザビリティ評価手法の中でインスペクション法に類別される認知的ウォークスルー法(CW 法)は,分析者自身がユーザの振舞いをシミュレートしながら一連の質問に回答することによって,システムとの円滑なインタラクションの妨げとなる問題点を洗い出す評価手法である.CW 法は漸次改良が加えられてきたが,実用性を重視して簡略化された第3 版では質問記述が抽象的で回答しにくいといった問題が指摘されている.本稿では,操作の対象と行為の区別および知覚と解釈の区別を導入してD.A. Norman の7 段階モデルを拡張した.その拡張モデルに基づき質問項目の見直しと明確化を図り,CW 法の改良を試みた.Web ユーザビリティ評価を題材として問題発見効率について比較実験を行った結果,評価対象の特性を考慮して自明な質問を省略することによって,提案手法は第3版と同等の回答時間でより高い問題検出率を示した.The cognitive walkthrough (CW) is a usability inspection method where analysts are asked to simulate the user's cognitive behavior and answer a series of evaluation questions for each step of a task. Negative answers indicate that such steps will be difficult to learn and thus likely to cause usability problems. The CW underwent a series of modifications to improve its applicability to real-world development processes. The current version, much simplified from its predecessors, has now only four evaluation questions. These questions, however, seem to be so abstract that they are difficult to answer properly unless one has expertise in cognitive science and/or usability evaluation. In this paper we propose a modified CW method whose evaluation questions are formulated based on an extended model of human-computer interaction that explicitly distinguishes between object and action, and perceiving and understanding. Applied to Web usability evaluation, the modified CW was shown to be more effective in identifying usability problems while remaining as efficient as the current CW.