1. The behaviour of individual roach was tested in a horizontal linear gradient of roach odour. Tests with a chemical model suggested that the gradient was logarithmic.
2. The swimming speed of the fish was not directly related to odour concentration but was low when fish swam into decreasing concentration and much higher in the opposite direction. This does not support an ortho-kinesis.
3. Turning was more frequent at the high concentration end of the gradient which the fish ‘preferred’. This is the reverse of klino-kinesis.
4. It is suggested that orientation is by spatial and temporal comparison of intensities, that speed is related to change in stimulation and that variation in turning serves to maintain fish at high odour concentrations.