The paper performs a panel regression on the definitionally uniform data now available for a five-year period ending in 1999-2000, on non-performing loans of commercial banks. The exercise is confined to 27 public sector banks, so as to investigate variations within a class that is homogenous on the ownership dimension. The exercise groups banks with higher than average NPAs into those explained by poor operating efficiency, and those where the operational indicator does not suffice to explain the high level of NPAs and leaves an unexplained intercept shift. Two of the three weak banks identified by the Varma Committee, Indian Bank and United Bank of India, fall in this category. Recapitalisation of these banks with operational restructuring may therefore not be the solution, since there is clearly a residual problem even after controlling for operating efficiency.