A critical review of dispersivity observations from 59 different field sites was developed by compiling extensive tabulations of information on aquifer type, hydraulic properties, flow configuration, type of monitoring network, tracer, method of data interpretation, overall scale of observation and longitudinal, horizontal transverse and vertical transverse dispersivities from original sources. This information was then used to classify the dispersivity data into three reliability classes. Overall, the data indicate a trend of systematic increase of the longitudinal dispersivity with observation scale but the trend is much less clear when the reliability of the data is considered. The longitudinal dispersivities ranged from 10−2 to 104 m for scales ranging from 10−1 to 105 m, but the largest scale for high reliability data was only 250 m. When the data are classified according to porous versus fractured media there does not appear to be any significant difference between these aquifer types. At a given scale, the longitudinal dispersivity values are found to range over 2–3 orders of magnitude and the higher reliability data tend to fall in the lower portion of this range. It is not appropriate to represent the longitudinal dispersivity data by a single universal line. The variations in dispersivity reflect the influence of differing degrees of aquifer heterogeneity at different sites. The data on transverse dispersivities are more limited but clearly indicate that vertical transverse dispersivities are typically an order of magnitude smaller than horizontal transverse dispersivities. Reanalyses of data from several of the field sites show that improved interpretations most often lead to smaller dispersivities. Overall, it is concluded that longitudinal dispersivities in the lower part of the indicated range are more likely to be realistic for field applications.