Temperature observations routinely taken over a period of years are analyzed for information on the space/time statistical structure of the short-term, climate-related variability in the western North Pacific. Observations were taken by the RV Ryofu Maru, making winter hydrographic transects along 137¿E from Japan to the equator each year from 1967--1974, and by the Japanese Far Seas Fisheries Agency, making 120 quasi-meridional bathythermograph sections between 125¿E and 180¿ from Japan to the equator during 1968--1982. Meridional wave number spectra subsurface temperature over the depth range 100--300 m are calculated from the sections. Spectra vary significantly between the tropical region south of 17.5¿N and the subtropical regions north of 17.5¿N. In both regions, wave number spectra increase in spectral energy density as the inverse square of the wave number from wavelengths of 200--1200 km; this indicates that the spectra derive from a first-order, autoregressive process. Decorrelation scales associated with the first-order process are very different for the two regions. Evaluation of these parameters is obtained by fitting the observed wave number spectra with a model spectrum that is composed of a first-order autoregressive process and a white noise process. The best-fit decorrelation scale of the first-order process is 600 km in the tropics (south of 17.5¿N) and 300 km in the subtropics (north of 17.5¿N). The model permits the partition of temperature variance into climate-related signal variance, and noise variance that cannot be resolved by routine monitoring. The ratio of noise to signal variance is 0.3 in the tropics and 1.0 in the subtropics. Estimates of the time and zonal space decorrelation scales are determined from the autocorrelation analysis of the Japanese Far Seas Fisheries Agency temperature data. Both north and south of 17.5¿N, the decorrelation time scale in the subsurface temperature is approximately 6 months. The zonal decorrelation length scale in the subsurface temperature is approximately 10¿ longitude for the tropical region south of 17.5¿N, but only about 2.5¿ longitude for the subtropical region north of 17.5¿N. For equal time scales, it is shown that the spectral shapes and the latitudinal differences in zonal space scale are consistent with the representation of upper ocean thermal variability in terms of nondispersive baroclinic long waves (i.e., Rossby waves). Discussed are implications of this space/time statistical information on the design of monitoring networks used to detect short-term climate-related variability in thermal structure.