Harmonic priming studies have repeatedly demonstrated that a tonal context primes listeners to expect a tonally related target chord (e.g., Bigand & Pineau, 1997). Tillmann and Bigand (2001) have shown, however, that scrambling the order of chords in the context fails to inhibit processing, suggesting temporal order only weakly contributes to harmonic priming effects. Given the recent claims emerging out of corpus studies of tonal harmony that temporal order is a fundamental organizing principle in many musical styles (e.g., Moss et al., 2019), this study replicates Tillmann and Bigand’s experimental design, but trains a probabilistic model on a corpus of chord annotations to select the scrambled conditions. Sixteen 9-chord sequences were selected from Bach’s chorales that either remain unchanged (thereby reflecting low temporal incoherence), or were scrambled to produce increasingly incoherent sequences (i.e., medium or high). To produce the scrambled conditions, a variable-order n-gram model (Pearce, 2005) was trained on a corpus of Roman numeral annotations from 100 Bach chorales. It then identified the scrambled versions that produced either medium or high estimates of model uncertainty. Sixty participants (30 musicians) indicated as quickly as possible whether the target chord was in or out of tune. In contrast to previous findings, correct response times from both groups replicated the ascending low-to-high staircase found in the model estimates. Nonmusicians also demonstrated an ascending low-to-high in-tune bias. To our knowledge, these data provide the first behavioral evidence that harmonic priming effects reflect the order of chords in a sequence.