Phytoplankton samples collected from two cruises off the coast of Chile were analysed for pigment composition and absorption characteristics. High pigment concentrations (up to 20 mg chl-a m-3) were found in the upwelled waters over the shelf break off the coast of Concepción during spring (October 1998), but relatively oligotrophic conditions were found further offshore. Similarly, stations further north (between Coquimbo and Iquique), sampled during the austral summer (February 1999), also showed low pigment concentrations, characterised by the presence of prymnesiophytes, and cyanobacteria including Prochlorococcus sp. The specific absorption coefficient of phytoplankton at 443 nm (a*ph(443)) was much higher for the offshore population than the inshore population, which was dominated by large diatoms. These differences are attributed to changes in pigment packaging and pigment composition. The relative proportion of non-photosynthetic carotenoids to chl-a, together with the ratio of the peak height of the Gaussian bands in the blue and red regions of the spectrum, p(435)/p(676), (an indicator of the importance of the packaging effect) could account for up to 92 % of the total variation in a*ph(443). Blue/green absorption ratios were strongly related to the relative concentration of 19'-hexanoyloxyfucoxanthin and fucoxanthin. A reasonable agreement was found between in situ and satellite estimates of chl-a (SeaWiFS data) despite the large variability in phytoplankton specific absorption coefficients, suggesting that the `global' absorption-to-chlorophyll relationships encompass the regional variations observed off the coast of Chile. Satellite chl-a was overestimated in oligotrophic water when compared to HPLC chl-a measurements, apparently because of the high specific absorption coefficients of phytoplankton in the offshore waters. On the other hand, ship and satellite data were in closer agreement when in situ fluorometric chl-a data was used. It is likely that the correlation between in situ and satellite chl-a could be improved by using regional algorithms