Speech graph analysis (SGA) of dreams has recently shown promise as an objective and language-invariant diagnostic tool that can aid neuropsychiatric diagnosis. Whilst the notion that dreaming mentations reflect distinct physiologic processes is not new, such studies in patients with sleep disorders remain exceptionally scarce. Here, using SGA and other dream content analyses, we set to investigate structural and thematic differences in morning dream recalls of patients diagnosed with Non-Rapid Eye Movement Parasomnia (NREMP) and Idiopathic REM Sleep Behavior Disorder (iRBD). A retrospective cross-sectional study of morning dream recalls of iRBD and NREMP patients was undertaken. Traditional dream content analyses, such as Orlinsky and Hall and Van de Castle analyses, were initially conducted. Subsequently, SGA was performed in order to objectively quantify structural speech differences between the dream recalls of the two patient groups. Comparable rate of morning recall of dreams in the sleep laboratory was recorded; 25% of iRBD and 18.35% of NREMP patients. Aggression in dreams was recorded by 28.57% iRBD versus 20.00% in NREMP group. iRBD patients were more likely to recall dreams (iRBD vs NREMP; P = 0.007), but they also had more white dreams, ie having a feeling of having dreamt, but with no memory of it. Visual and quantitative graph speech analyses of iRBD dreams suggested stable sequential structure, reflecting the linearity of the chronological narrative. Conversely, NREMP dream reports displayed more recursive, less stable systems, with significantly higher scores of graph connectivity measures. The findings of our exploratory study suggest that iRBD and NREMP patients may not only differ on what is recalled in their dreams but also, perhaps more strikingly, on how dreams are recalled. It is hoped that future SGA-led dream investigations of larger groups of patients will help discern distinct mechanistic underpinnings and any associated clinical implications.