The discrimination of harmful algal blooms (HABs) from space would benefit both the capability of early warning systems and the study of environmental factors affecting the initiation of blooms. Unfortunately, there are no published techniques using global monitoring satellite sensors to distinguish the resulting subtle changes in ocean colour, so in situ sampling is needed to identify the species in any observed bloom. This paper investigates multivariate classification as an objective means to discriminate harmful and harmless algae and monitor their dynamics using ocean colour data derived from satellite sensors. The classifier is trained and tested using Sea‐viewing Wide Field‐of‐view Sensor (SeaWiFS) data, though the method is designed to be generic for other sensors. Time‐series results are presented using the new HAB likelihood index and suggest that SeaWiFS has some capability for observing the dynamic evolution of harmful blooms of Karenia mikimotoi , Chattonella verruculosa and cyanobacteria. Further, a multi‐band spatial subtraction algorithm is described to automate the identification of bloom regions and improve the accuracy in discriminating HABs.