Abstract The process of measurement is intrinsically error prone. One form of error, traditionally conceptualized as a random variable added to the true score and commonly called measurement error, has been studied extensively, both as a component of responses and as a characteristic of the environment. In this paper we are interested in error in the environment, or what Brunswik (1956) called “ecological unreliability.” Specifically we are interested in a form of error in which, on some proportion of trials, the obtained measurement is completely unrelated to what is supposedly being measured. This form of error, defined and labeled herein as “system failure” (SF) error, has not been commonly studied in investigations of probabilistic environments. A simple application of the Pearson product moment correlation is derived and suggested as a means of making experimental environments with SF error partly commensurable with those in which error is traditionally manipulated. The relation between the proportion of SF errors and ϱ is shown for a bivariate case and for a simple multivariate case. Monte Carlo estimates of the sampling distribution of r in environments with SF error are provided for four levels of SF error for one example of an experimental environment.