Colour data always has a limited accuracy, which is caused by two separate effects. Experimental random error is induced by the actual measurement. Quantisation error results from the quantised representation of the data. Both error types propagate through the processing, while quantisation in intermediate steps can cause additional error. Existing methods deal only with the propagation of random error, or with the effect of initial quantisation.We study the combined effect of both error types and arrive at the result that their error variances can be added. For the case of successive quantisations, we propose a computational method for calculating the output error. We illustrate this by numerical data for various transforms.