Three criteria are used to find the best of eight directional symbols suitable for traffic control signals: visibility when blurred, visibility at low contrast and identification in relation to acceptability. These are used instead of the more conventional field type criteria of recognition distance and reaction time. The results of the three experiments show that: 1. the rank order of performance with different symbols differs with the criterion used, 2. the blur variable, which is a relatively new visibility criterion, has proved successful in differentiating between symbols, and 3. visibility can be better for unconventional symbols than for the more conventional forms.