A model for the attenuation of high-frequency (1-8 Hz) P waves is developed for paths from eastern Kazakhstan to arrays in Scotland (EKA), Canada (YKA), India (GBA), and Australia (WRA). The attenuation model includes contributions from both intrinsic absorption and scattering. Distinction between the two is especially important for forward modeling because the dispersion is quite different. However, no satisfactory complex operator for representing the scattering attenuation appears to be available at present. Attenuation consistent with an absorption band model dominates at low frequencies and is strongly dependent on frequency in this band. Attenuation that appears to be due to scattering has an important effect above 2.5 Hz and dominates at high frequency. It can be represented by a constant effective Q-x, or by Q-• • f The details of the best fitting model depend on which of these is chosen and on the value of to* (travel time/Q ratio at long period) which is poorly constrained by these data. The general conclusion is that the intrinsic attenuation can be represented by an absorption band model with 0.04 _• z,• _• 0.1, while 0.5 _• to* -• 1.0. Superimposed on this is scattering attenuation, which can be represented by a constant t* of about 0.1 s or by t* = Af, with A about 0.01. Some differences are seen among the four paths, with the greatest attenuation in the 1-3 Hz band seen at YKA and the least at EKA. Above 3 Hz the attenuation effects are essentially identical for all four paths. mori and Anderson, 1977] provides a good theoretical framework for interpreting attenuation data and is now used in most studies of frequency dependent attenuation. However, this model does not represent the effects of scattering which can be an important cause for amplitude attenuation at high frequencies. Thus the nature of the attenuation mechanism is an impoi'tant issue in the short-period band.The objective of this study is to determine the attenuation of teleseismic P waves in the 1-8 Hz band for several paths for which particularly good data are available. The technique is to analyze the spectra of very short time windows chosen to isolate the first-arriving P wave pulse. The attenuation effects of interest are those which influence the amplitude of this pulse and therefore rnb. The attenuation determined this way is not necessarily the same as that determined from a longer window including much of the P coda.The main reason the frequency dependence near 1 Hz remains so poorly understood is that adequate data are difficult to obtain. The illuminating characteristics are obscured by noise and by complexity caused by path effects. Also, there is a strong trade-off between source and attenuation effects, since most of the events that provide the best data have corner frequencies near 1 Hz. In this study these deficiencies are largely overcome by using recordings of Soviet Eastern Kazakhstan explosions from four arrays designed by the United Kingdom Atomic Energy Authority and operated by organizations in Scotland, Canada...