A number of studies aimed to distinguish the structural patterns at the interfaces of obligate and non-obligate protein-protein complexes. These studies revealed better geometric complementarity of protomers in obligate complexes over non-obligates. We showed that protein surface roughness can be used to explain this observation. Using smoothened atomic fractal dimension (SAFD) as a descriptor, this work investigates the role of interface roughness in the molecular recognition of these two types of protein-protein complexes. We studied 52 obligate and 62 nonobligate heteromeric high quality crystal structures from benchmark data sets. We found that distribution of interface roughness values obligate and non-obligates are quite similar. However, we observed a distinct preference for obligate protomers to complex with chains having similar roughness. The roughness pairing is correlated in obligates only. The later indicates, an increase/decrease of roughness in one chain causes a proportional change in roughness in its binding partner. Based on these observations we proposed that similar and correlated roughness pairing leads to more interdigitation and contacts at the interface leading to better geometric fit in obligates. We propose that roughness information can find useful application in improving machine learning based complex type classifiers and filtering protein-protein docking solutions.