The S stars are late-type giants whose spectra resemble those of M giants, with the addition of distinctive molecular bands of ZrO (Merrill 1922). The presence of ZrO bands is often considered as a direct consequence of molecular equilibrium in the special circumstances where the atmospheric C/O ratio is within 10% of unity (e.g. Scalo & Ross 1976). However, Piccirillo (1980) has shown that the above statement is only valid for stars with T < 3000 K. At higher temperatures, an enhanced Zr abundance, rather than a C/O ratio close to unity, is the dominant factor in the development of strong ZrO bands. Detailed abundance analyses (Smith & Lambert 1990) have shown that the envelopes of S stars are enriched in heavy elements like Zr, and so bear the signature of the s-process of nucleosynthesis (Käppeler et al. 1989). Although larger than in M giants, the C/O ratio of S stars is not necessarily close to unity (Smith & Lambert 1990), except in the so-called SC stars (Dominy et al. 1986).
When S stars were still believed to be objects with C/O close to unity, they were naturally considered as transition objects between M giants (C/O < 1) and carbon stars (C/O > 1) on the asymptotic giant branch (AGB) (Iben & Renzini 1983). Support to this scenario is provided by observations of S stars on the upper AGB of globular clusters in the Magellanic Clouds (Bessell et al. 1983; Lloyd Evans 1983-1985). In this evolutionary phase, low- and intermediate-mass stars are characterized by a double (H, He) burning-shell structure which is thermally unstable. The thermal instabilities ("thermal pulses'') developing in the He-burning shell are the site of a rich nucleosynthesis (Frost & Lattanzio 1995), probably including the s-process, although its detailed mode of operation remains poorly understood (e.g. Sackmann & Boothroyd 1991; Herwig et al. 1997). In the receding phase of the thermal instability, the convective outer envelope may plunge ("third dredge-up'') into the intershell zone containing the He-burning ashes, and bring fresh carbon and s-process elements to the surface.
However, several observations have challenged this traditional M-S-C evolution sequence. The first set of observations relates to Tc, an element with no stable isotopes, discovered in the spectra of some S stars by Merrill (1952). If the s-process indeed occurred during recent thermal pulses in S stars, Tc should be observed at the surface along with the other s-process elements (Mathews et al. 1986). Little et al. (1987) found however that only long-period Mira or semiregular S stars (i.e., intrinsically bright S stars) exhibit Tc lines. Second, the broad range of IRAS colors exhibited by S stars (e.g., Jorissen et al. 1993) is difficult to reconcile with the idea that they represent a brief transition phase as the star evolves from an oxygen-rich M giant with C/O < 1 into a C star with C/O > 1. M and C stars occupy well-defined regions in the IRAS color-color diagram, and it is not clear how S stars fit into the (much debated) evolutionary sequence joining M stars to C stars in that diagram (Willems & de Jong 1986, 1988; Chan & Kwok 1988; Zuckerman & Maddalena 1989; de Jong 1989).
These problems received a new impetus with the discovery that the barium stars, a family of peculiar red giant (PRG) stars of spectral type G and K, are all members of binary systems (McClure et al. 1980; McClure 1983). Iben & Renzini (1983) were the first to propose that Tc-poor S stars could perhaps be the cooler analogs of the barium stars. Long-term radial-velocity monitoring confirmed this suggestion, and it is now clear that Tc-poor S stars are binary stars (Smith & Lambert 1988; Brown et al. 1990; Jorissen & Mayor 1992; Jorissen et al. 1993; Johnson et al. 1993) with orbital elements identical to those of barium stars (Jorissen et al. 1998). Tc-poor S stars are now referred to as "extrinsic S stars'', because, like barium stars, they owe their chemical peculiarities to mass transfer across the binary system. On the contrary, Tc-rich, "intrinsic S stars'' are genuine thermally-pulsing (TP) stars on the TP-AGB. Since the C/O ratio of extrinsic S stars depends on the details of the mass accretion process, it is not necessarily close to unity, but as discussed above, neither abundance analyses nor predictions of molecular chemical equilibrium really require C/O to be close to unity in S stars.
In Paper I (Jorissen et al. 1993; see also Groenewegen 1993), it was shown that the correlation Tc-poor/binary found for S stars could be extended to their IRAS colors, since another distinctive property of binary, Tc-poor S stars is the absence of IR excesses. Their IRAS colors simply reflect the photospheric colors, contrary to Tc-rich S stars which usually exhibit IR excesses. These IR excesses are caused by circumstellar dust, and are indicative of substantial mass loss, thus suggesting that Tc-rich S stars are more massive and/or more evolved than Tc-poor S stars.
The possibility offered by the circumstellar properties of S stars to probe their
evolutionary status has been used in several recent studies
(Jura 1988;
Chen & Kwok 1993;
Bieging & Latter 1994;
Sahai & Liechti 1995). However, these studies
still rely on hypotheses not fully consistent with the dichotomy recently found
among S stars, as discussed above, and their conclusions may therefore be somewhat
biased. For example, it makes no sense to test the AGB evolutionary sequence
M-S-C using a sample of S stars not properly cleaned from its extrinsic content.
In addition,
models inferring the chemical nature (carbonaceous or silicate) of the dust
grains from models assuming that C/O is close to unity in the photosphere do not
sample the whole parameter space occupied by these stars.
A similarly incorrect corollary consists of concluding that dust production, and
thus mass loss, is not very efficient in S stars because dust-seed molecules no
longer form in the absence of any free C or O atoms,
since these are all tied up in CO when C/O 
.
It is the purpose of this paper to rediscuss the circumstellar properties of S stars and to put these properties in perspective with our current understanding of the evolutionary status of S stars, in particular the intrinsic/extrinsic dichotomy. An extensive data set probing the circumstellar environment of S stars (IRAS colors, maser emission, CO rotational lines) has therefore been collected and critically evaluated. This data set combines new material with existing results collected from the literature.
The IRAS colors of S stars provide a first way to probe their circumstellar
environment, and more specifically, to evaluate the amount and nature of the dust
surrounding these objects. A re-evaluation of the flux
densities listed in the IRAS Point Source Catalogue (IRAS
Science Team 1988; PSC) was necessary to
circumvent the problems of interpretation
related to the intrinsic IR variability of these sources and
to their possible contamination by Galactic cirrus emission. These effects are
not always properly handled in the PSC. This re-evaluation was performed by
co-adding the raw scans (Sect. 2.2). The clean flux densities were then used to
define five regions in the (K-[12], [25]-[60]) color-color diagram
(Sect. 2.3) which contain stars of similar extrinsic or intrinsic nature
and of similar ZrO/TiO, C/O and IR spectral indices.
Several S stars with shells resolved at 60 (and sometimes 100) 
m have been
found by comparing the source profile with the IRAS point source response
function (Sect. 2.4).
Resolved shells at 60 m appear to be correlated with large 60 m
excesses, suggesting that these shells have detached from their parent star.
A simple model of the dust shell has been used to predict its IR colors and to
infer the chemical nature of the dust grains (Sect. 4), using constraints provided
by the detection or non-detection of maser emission (Sect. 3).
Finally, mass loss rates have been derived in an homogeneous way (Sect. 5.3) from
the intensities of the CO millimeter-wave
lines, derived from new observations with the
Caltech Submillimeter Observatory (Sect. 5.1) or collected from the literature
(Sect. 5.2). The mass loss rates, wind expansion velocities, IR colors and
extrinsic/intrinsic nature are then discussed together in Sect. 6.