The tacit-knowledge approach relies on a critical-incident technique to identify examples of tacit knowledge acquired in solving real-word problems. That is, we interview domain experts to identify incidents that reflect important learning lessons, and ask them to express in their own words the knowledge gained from those situations. We do not rely solely on the individuals who provided the incidents to determine which items of knowledge are more or less effective. We use subsequent analyses to identify the items that are "critical" to performance.

The tacit-knowledge approach shares with the simulation approach the view that measuring practically relevant behavior in a test situation depends, in part, on the extent to which the task resembles those tasks found in everyday life. As such, we attempt to include sufficient detail in our measure to provide respondents with a realistic picture of the situation. However, we have relied primarily on a paper-and-pencil format to present this information rather than simulations for reasons of practicality, with the exception of our tacit-knowledge-acquisition task for sales (Sternberg et al., 1993). We have chosen to provide better coverage of the performance domain at the potential cost of lower fidelity. Future testing, however, is moving in the direction of more performancebased, high-fidelity assessment.

The tacit-knowledge approach is linked most closely to that of situational-judgment testing. We present situation descriptions, often based on actual situations of position incumbents, followed by several possible responses to those situations. The number of response options range between five and twenty. Individuals are asked to rate on a Likert scale the quality or appropriateness of each option for addressing the problem presented in the situation.

For example, in a hypothetical situation, an administrative assistant realizes that there is a factual error in a memo her boss has written and the memo needs to be sent out immediately. The boss is in a closed-door meeting. The respondent is asked to rate several options (usually on a 1 = low to 9 = high scale) for solving the problem. Examples of responses include (a) interrupting the meeting to show the boss the error, (b) fixing the error oneself and sending out the revision, and (c) fixing the error but waiting to send out the memo until the assistant can run it by the boss.

The set of ratings the individual generates for all the situations is used to assess the individual's tacit knowledge for that domain. Similar to SJTs, the scoring of tacitknowledge tests often rely on the judgments of experts. In general, tacit-knowledge tests have been scored in one of three ways: (a) by correlating participants' responses with an index of group membership (i.e., expert, intermediate, novice), (b) by judging the degree to which participants' responses conform to professional "rules of thumb," or (c) by computing the difference between participants' responses and an expert prototype. To understand better what tacit-knowledge tests are designed to measure, we consider tacit knowledge as a measurement construct.