Threshold models indicate that the best way to improve the effectiveness of sensory screening of water at the point of production is to give assessors sensory training, but we argue that it is likely to be ineffective. The notion of sensory threshold has been displaced in psychophysics by signal-detection models which assume that the assessor is uncertain on every trial, but makes an unconscious decision about the presence or absence of the attribute. The decision can be influenced by many factors other than the strength of the attribute and the sensitivity of the assessor. Taking account of possible changes in response criterion as well as in sensory sensitivity indicates a quite different way of improving sensory screening by attempting to counteract vigilance decrement and the effects of relative judgement. This approach indicates that, rather than trying to increase the sensitivity of assessors, we should aim to prevent shifts in response criterion and the transmission of errors from one sample to the next. A method that has proved successful with industrial inspection and military watchkeeping is to add artificial 'signals' to those being assessed and give the assessors immediate feedback on their success in judging them.