The gulf of Cariaco, Venezuela, is a rather isolated piece of coastal ocean where spatio-temporal variability has been measured. In 1986, three repeated surveys have been carried out in this gulf. The important differences in biomass estimates are attributed to sampling error only. The biological nature of this variability is studied. It is due to a high concentration of the biomass in a few schools and correlatively to the hit or the miss of these large schools, during the sampling. Structural information for predicting them is sought. Also, the number of schools per elementary sampling distance unit (ESDU) and the biomass in each school were measured by echointegration. It is found that there is no clear relation between these two variables. Thus their spatial structures are studied separately, using geostatistics. The number of schools per ESDU is structured in space. Its histogram is not very dissymmetrical. The error variance of its mean estimate is reasonable. The biomass per school is not structured in space. Its histogram is highly skewed. The error variance of its mean estimate is high. The major error variance term on the abundance is due to the imprecision on the estimate of the mean biomass per school. The major quantitative problem is the indeterminancy at the sampling stage of the maximum biomass per school and correlatively the major problem is the characterization of the skewness of the biomass per school histogram. The possibility of using the number of schools as an index of biomass is discussed.