Harmful algal blooms (HAB) can lead to great economic losses to fisheries and significant adverse impacts on the environment. And yet the onset of HAB, a worldwide problem, is notoriously difficult to predict. This paper describes the design, development, and initial operation of a real time, remotely controlled, early warning system for algal blooms and red tides in a coastal field research station. The system measures solar radiation, wind velocity, tidal level and velocity, dissolved oxygen and chlorophyll fluorescence at three depths, supplemented by regular onsite sampling for subsequent chemical analysis of nutrient and chlorophyll concentrations, and taxonomic examination. An example of successful detection and monitoring of an algal bloom event (and also a red tide) is given. Preliminary results of modelling chlorophyll concentrations using artificial neural networks (ANN) are also presented. Different network structures are trained on a six year biweekly water quality data set, and tested on an independent 3 year data set. Unlike previous work on limnological and riverine applications, the results show that rather good predictions of long term trends in algal biomass can be obtained using only Chlorophyll-a and (or without) total inorganic nitrogen (time delayed for one week) as input nodes. On the other hand, the phase error of predictions renders the ANN method unsuitable for short term forecasts of algal blooms - which can take off and collapse in 7-10 days.