A three-component model of ocean colour is presented, that takes into account contributions of phytoplankton, non-chlorophyllous particles and dissolved organic matter. The model is based on theoretical considerations and is validated by comparison with observed reflectance spectra. It is then used to address the problem of estimating chlorophyll concentration in coastal (case 2) waters by remote sensing. Principal component analysis is carried out on a large number of reflectance spectra simulated using the model. The results indicate the possibility of chlorophyll retrieval in at least some case 2 waters. Retrieval becomes difficult in waters where the signal from small quantities of chlorophyll is drowned in the noise from large quantities of non-chlorophyllous particles and dissolved organic matter. Non-chlorophyllous particles are the most easy to recover, while dissolved organic matter appears to be the most difficult. The non-linearity of the system limits the possibility of a single algorithm to cover all ranges of variation. It appears possible to pass from full spectral data to five selected wavelengths without any loss in the retrieval efficiencies. The use of 400 nm-channel data in the algorithm, in addition to the data from the usual 440, 520 and 550nm channels, considerably increases the possibility of distinguishing phytoplankton from dissolved organic matter.