Young infants have demonstrated a remarkable sensitivity to probabilistic relations among visual features (Fiser & Aslin, 2002; Kirkham et al., 2002). Previous research has raised important questions regarding the usefulness of statistical learning in an environment filled with variability and noise, such as an infant's natural world. In an eye-tracking experiment, 8-month-old infants viewed sequences of spatio-temporal events with three different transitional probabilities (1.0-Deterministic, 0.75-High probability, and 0.5-Low probability). Across two between-subjects conditions, the sequences were presented with or without competing visual distracters. Results show that as transitional probability decreased, infants distributed less attention to the predictable locations and their anticipations were less often correct. With no distraction, infants had faster saccadic latencies to the high probability events; however, with distracters present in the stimulus environment, infants' eye movements shifted to favour the deterministic relations. These findings suggest that infants integrate multiple sources of variability to guide visual attention and facilitate the detection and learning of statistically reliable events.