An objective analysis technique is used to design an optimal oceanographic network for the eastern Pacific Ocean from 160¿W--80¿W, 20¿S--20¿N. Expendable bathythermograph observations from a sequence of transequatorial transects (1979--1984) were used to estimate temporal and meridional mean and correlation statistics, while bathy messages in the global telecommunications system (1979--1982) were used to calculate zonal statistics. The variables sea-surface temperature and depth of 20 ¿C isotherm are presented here as representing the thermal structure of both the surface and subsurface layers. In the eastern Pacific, the thermal temporal and spatial signals fluctuate according to the presence and strength of El Ni¿o. During ENSO it is not uncommon for these scales to double in value. To resolve characteristically shorter signals during the typical years, data associated with the larger scales of ENSO events must be excluded. Scales of variability in the thermal field were also determined to be very different in five latitudinally delineated regions corresponding to the major equatorial current system of the eastern Pacific region. In general, space and time scales representative of the subsurface structure are smaller than scales of sea-surface temperature and are consequently the limiting factor in network design. In the design of an XBT network for the entire tropical eastern Pacific, decorrelation scales of 3¿ latitude, 15¿ longitude and 2 months are determined as being most representative and capable of resolving the subsurface thermal signal. Sub-regional inhomogeneity of signal-to-noise ratio for the eastern Pacific (ranging from 0.5--1.5) implies that a uniform scale is not appropriate especially for quantitative studies of mapping error. Structure of interpolation errors suggest that optimal sampling is achieved with two to three samples per decorrelation scale i.e. one XBT station every 1¿--1.5¿ latitude and 5¿--7.5¿ longitude. ¿American Geophysical Union 1991