Methods to reduce intrusive memories (e.g., of traumatic events) should ideally spare voluntary memory for the same event (e.g., to report on the event in court). Single-trace memory accounts assume that interfering with a trace should impact both its involuntary and voluntary expressions, whereas separate-trace accounts assume these two can dissociate, allowing for selective interference. This possibility was investigated in 3 experiments. Nonclinical participants viewed a trauma film followed by an interference task (Tetris game-play after reminder cues). Next, memory for the film was assessed with various measures. The interference task reduced the number of intrusive memories (diary-based, Experiments 1 and 2), but spared performance on well-matched measures of voluntary retrieval—free recall (Experiment 1) and recognition (Experiments 1 and 2)—challenging single-trace accounts. The interference task did not affect other measures of involuntary retrieval—perceptual priming (Experiment 1) or attentional bias (Experiment 2). However, the interference task did reduce the number of intrusive memories in a laboratory-based vigilance-intrusion task (Experiments 2 and 3), irrespective of concurrent working memory load during intrusion retrieval (Experiment 3). Collectively, results reveal a robust dissociation between intrusive and voluntary memories, having ruled out key methodological differences between how these two memory expressions are assessed, namely cue overlap (Experiment 1), attentional capture (Experiment 2), and retrieval load (Experiment 3). We argue that the inability of these retrieval factors to explain the selective interference is more compatible with separate-trace than single-trace accounts. Further theoretical developments are needed to account for this clinically important distinction between intrusive memories and their voluntary counterpart.