Chronic low back pain is a common and heterogeneous disease defined by sensory and affective experiences. In pre-registered analyses, we examined variability in within-person relationships between negative affect (NA) and physical pain over time using multi-level dynamic structural equation modeling. We further examined how this variability relates to risk for future opioid-related problems as assessed by the Screener and Opioid Assessment for Patients with Pain-Revised. Patients with chronic low back pain (N = 87) completed an average of 94 Ecological Momentary Assessment surveys each over two weeks. In the group-level model, lagged pathways suggested that increased pain predicted increased NA two hours later (β = 0.10, p < .001), and increased NA predicted increased pain two hours later (β = 0.16, p < .001) for individuals on average. However, there was significant variability in these effects, such that NA significantly predicted pain two hours later for 48% of participants, and pain significantly predicted NA two hours later for 8% of participants. In a multiple regression analysis, risk for future opioid-related problems was positively associated with the prospective relationship of pain predicting increased NA two hours later (b = 8.05, p = .007) and the individual-level correlation between pain and NA (b = 1.99, p = .002). These results suggest that variability in within-person symptom dynamics may help identify chronic pain patients who are at greater risk of opioid-related problems. Future research should examine whether personalizing treatment based on relationships between pain and affect can mitigate opioid-related risks.