This paper is part of an on-going empirical research programme to develop an improved understanding of the implementation and evaluation of on-line diagnostics in software. In this study we have concentrated on the hypothesis that residual design errors exist because their coupling to the input space is very small, making them difficult to detect in normal testing. The objective of the reported experiment was basically to add a simple group of diagnostic checks to a reasonably complex program and use arbitrary fault injection to assess the error detection in relation to the coupling of the fault to the input space. The results were promising in that they demonstrated no significant deterioration in the effectiveness of the diagnostics as the fault coupling to the input space decreased. On this basis the use of diagnostics can be seen as supplementary to validation testing.