Phonological contrasts in speech production and perception are frequently studied with acoustic and/or auditory cues. Less is known about how visual cues are incorporated in a contrast, and how these cues are affected by an acoustic merger-in-progress, especially for changes that can be initiated by different articulatory gestures. This paper investigates the acoustic and visual cues in the ongoing Mandarin sibilant merger, where retroflexes are merging with alveolars. We analyzed audiovisual production data using Computer Vision articulography and found that speakers range from having an almost complete visual merger to a complete visual distinction. The direction of the visual cue merger was also consistent with the direction of the acoustic merger. Further, visual cues were found to be correlated with acoustic cues and strongly predicted spectral moments. We extend previous findings of cue weighting to a shared domain of audio and visual cues, and we discuss implications for the directionality of sound change.