Step length asymmetry (SLA) is common in most stroke survivors. Several studies have shown that factors such as paretic propulsion can explain between-subjects differences in SLA. However, whether the factors that account for between-subjects variance in SLA are consistent with those that account for within-subjects, stride-by-stride variance in SLA has not been determined. SLA direction is heterogeneous, and different impairments likely contribute to differences in SLA direction. Here, we identified common predictors between-subjects that explain within-subjects variance in SLA using sparse partial least squares regression (sPLSR). We determined whether the SLA predictors differ based on SLA direction and whether predictors obtained from within-subjects analyses were the same as those obtained from between-subjects analyses. We found that for participants who walked with longer paretic steps paretic double support time, braking impulse, peak vertical ground reaction force, and peak plantarflexion moment explained 59% of the within-subjects variance in SLA. However the within-subjects variance accounted for by each individual predictor was less than 10%. Peak paretic plantarflexion moment accounted for 4% of the within-subjects variance and 42% of the between-subjects variance in SLA. In participants who walked with shorter paretic steps, paretic and non-paretic braking impulse explained 18% of the within-subjects variance in SLA. Conversely, paretic braking impulse explained 68% of the between-subjects variance in SLA, but the association between SLA and paretic braking impulse was in the opposite direction for within-subjects vs. between-subjects analyses. Thus, the relationships that explain between-subjects variance might not account for within-subjects stride-by-stride variance in SLA.