Several computational models explaining fixation durations in scene viewing (Nuthmann, Smith, Engbert, & Henderson, 2010) and in reading (Engbert, Nuthmann, Richter, & Kliegl, 2005; Reichle, Pollatsek, Fisher, & Rayner, 1998) assume that saccade programming is completed in two stages: an initial, labile stage that is subject to cancellation and an subsequent, non-labile stage in which the program can no longer be cancelled. This distinction is motivated by findings from double-step experiments that used much simpler situations than scene viewing or reading. Here, we adopt a classic double-step paradigm to a scene-viewing context. In a Static condition targets are presented to the left or right of a central fixation cross along a horizontal axis while in a Scene condition targets are presented in a gaze contingent manner along a trajectory defined by the location of recent fixations. We found evidence in support of the claims that saccade cancellation occurs within a naturalistic scene-viewing context and that saccade cancellation can account for increases in observed fixation duration distributions. The duration of the non-labile stage was estimated to be longer in the Scene condition compared to the Static condition.