["Autism Research, EarlyView. ", "\nABSTRACT\nSleep disturbances are highly prevalent in autism spectrum disorder (ASD), yet conventional single‐night polysomnography (PSG) provides limited insight into dynamic sleep regulatory processes. Using the Synaptic Homeostasis Hypothesis as a conceptual framework, we quantified night‐to‐night changes (Δ; Night 2 − Night 1) in sleep architecture across two consecutive nights of home‐based PSG and examined their association with ASD symptom severity. Seventy‐nine children with ASD, stratified into mild ASD (M‐ASD, n = 35) and moderate‐to‐severe ASD (MS‐ASD, n = 44) subgroups using the Childhood Autism Rating Scale, and 30 typically developing controls (TDC) participated. Overall sleep‐continuity changes were broadly comparable between ASD and TDC groups. However, sleep‐architecture Δ measures showed a severity‐related pattern: the M‐ASD subgroup exhibited an overnight increase in rapid eye movement sleep percentage (ΔREM% = +2.6%, p = 0.002), whereas the MS‐ASD subgroup showed no significant change (ΔREM% = −0.1%, p = 0.922). An adjusted pairwise comparison indicated a greater ΔREM% increase in M‐ASD than in MS‐ASD (adjusted mean difference = 3.4%, 95% CI: 1.2 to 5.6; Bonferroni‐corrected p = 0.009). A similar exploratory difference was observed for NREM1%. ΔREM% was modestly and negatively associated with ASD symptom severity in correlation analysis (ρ = −0.234, p = 0.038) and age‐ and gender‐adjusted regression analysis (β = −0.286, p = 0.016). Sensitivity analyses showed broadly consistent patterns, although severity‐stratified overall effects were attenuated after FDR correction. These findings support further investigation of PSG‐derived dynamic sleep measures as hypothesis‐generating correlates of clinical heterogeneity in ASD.\n"]