Consistent with the Park and Levine's (PL) probability model of deception detection accuracy, previous research has shown that as the proportion of honest messages increases, there is a corresponding linear increase in correct truth–lie discrimination. Three experiments (N = 120, 205, and 243, respectively) varied the truth–lie base rate in an interactive deception detection task. Linear base‐rate effects were observed in all 3 experiments (average effect r#x02009;= .61) regardless of whether the judges were interactive participants or passive observers, previously acquainted or strangers, or previously exposed to truths or lies. The predictive power of the PL probability model appears robust and extends to interactive deception despite PL's logical incompatibility with interpersonal deception theory.