Many types of coherent radar spectra have a width in Doppler velocity units that is less than the ion acoustic speed of the medium. In spectra labeled as type 1 the mean Doppler shift of these narrow spectra matches the ion acoustic speed of the medium. There also exist narrow high-latitude spectra for which the mean Doppler shift is either markedly less or markedly more than the ion acoustic speed. We propose that electron density gradients with scale lengths as small as 100 m are at the origin of a large fraction of these narrow spectra near 50 MHz. The sharp density gradients in that case are created in regions of discrete auroral precipitation associated either with multiple narrow arcs or with sharp edges of boarder features. Using the same principle at radar frequencies in the 10- to 20-MHz range, we find that gradient scales from 20 to 30 km in size create a combination of fast and slow phase velocities closely resembling the spectral characteristics expected from NO+ ion cyclotron waves. However, gradients are not always responsible for slow narrow spectra; a detailed analysis of available observations has led us to conclude that the high-latitude E region cannot always be considered as fully turbulent even when appreciable coherent echo returns are registered by the radars. In particular, slow narrow spectra at 50 MHz are at times produced under gradient-free weakly turbulent conditions. In addition, at lower radar frequencies (10 to 20 MHz) the narrow spectral width of slowly moving waves and the morphology of these waves both suggest that the irregularities are generated indirectly via mode coupling of linearly unstable modes and that these ''secondary'' waves are themselves not coupling efficiently. This implies that processes other than mode coupling are contributing to the overall wave energy budget. In that case we suggest that the convective properties of the slowly growing modes are an important factor in removing wave energy, even for waves as small as few meters in wavelength. We also propose that there may be two distinct generation mechanisms for secondary waves at 10 MHz, each with its own mean Doppler shift behavior. ¿ American Geophysical Union 1994