Ross Perot was a relatively viable third party candidate in the 1992 US presidential election, but he was not any more in the 1996 election. This provides a good opportunity to analyze strategic voting behavior–voting for a candidate not most preferred by the voter–in the US presidential elections with panel data drawn from NES (National Election Studies). First, the 1992 election is analyzed with multinomial choice estimators. Second, using the estimates, each individual’s choice is predicted for the 1996 election. Third, those who were predicted to vote for Perot in 1996 but did not are identified as strategic voters and their profile is drawn. In addition to the main task of analyzing the strategic voting behavior, this paper does two additional tasks. First, analyzing the 1992 data with multinomial choice estimators, it is found that the following variables mattered significantly for the US presidential election: respondent and candidate ideology, personal finance, age, education, income, sex, abortion stance, health insurance policy, and welfare program policy. Second, critical mistakes in the literature in applying multinomial probit to election data are pointed.