We find that AGB stars separate in the 25–12 vs. 12-K color-color diagram according to their chemistry (O, S vs. C) and variability type (Miras vs. SRb/Lb). While discrimination according to the chemical composition is not surprising, the separation of Miras from SRb/Lb variables is unexpected.

We show that “standard” steady-state radiatively driven models provide excellent fits to the color distribution of Miras of all chemical types. However, these models are incapable of explaining the dust emission from O-rich SRb/Lb stars. The models can be altered to fit the data by postulating different optical properties for silicate grains, or by assuming that the dust temperature at the inner envelope radius is significantly lower (300–400 K) than typical condensation temperatures (800–1000 K), a possibility which is also supported by the detailed characteristics of LRS data. While such lower temperatures are required only for O- and S-rich SRb/Lb stars, they are also consistent with the colors of C-rich SRb/Lb stars.

The absence of hot dust for SRb/Lb stars can be interpreted as a recent (order of 100 yr) decrease in the mass-loss rate. The distribution of O-rich SRb/Lb stars in the 25–12 vs. K-12 color-color diagram shows that the mass-loss rate probably resumes again, on similar time scales. It cannot be ruled out that the mass-loss rate is changing periodically on such time scales, implying that the stars might oscillate between the Mira and SRb/Lb phases during their AGB evolution as proposed by Kerschbaum et al. (1996). Such a possibility appears to be supported by recent HST images of the Egg Nebula obtained by Sahai et al. (1997), the discovery of multiple CO winds reported by Knapp et al. (1998), and long-term visual light-curve changes detected for some stars by Mattei (1998).