During the course of the in-orbit calibration of the Short Wavelength
Spectrometer (SWS) it became apparent that its performance could be greatly
enhanced by an improved pointing of ISO - a typical Absolute Pointing Error
(APE) of 4 
was seen to limit the photometric calibration accuracy to about
30% and affect the wavelength calibration because of an induced effective
shift of one grating scanner step.
Because of the excellent stability of the ISO spacecraft, mainly due to
lower than expected thermoelastic distortions between the Star Tracker (STR)
and the optical axis of the telescope, a search for systematics in the APE was
carried out. This was followed by various fine-tunings to the system, which
culminated in a measured improvement to about 2 . Indications are that this
will be lowered to 
1 by next Spring. It is worth reporting on how
this was reached, considering that a major satellite system has been tuned
considerably beyond its specification (the APE spec was < 11.7 , 2-sigma,
half cone).
The first experiment, designed by the SWS Instrument Dedicated Team (IDT),
performed and analysed in collaboration with the CAM IDT, was carried out in
revolution 137. ISOCAM observed 79 stars, taken from the Ground Based
Preparatory Programme, which constitutes a pre-selected set of point-like
stars. The smallest pixel field of view (1.5 ) was used, in the LW9 band
(14-16 
m). The results indicated a possible trend of the APE with the
position of the Guide Star used for each of those pointings on the Star
Tracker (STR) CCD, but data was too noisy to come to a pattern for on-board
correction of the pointing algorithm parameters.
A refined version of the experiment was carried out in revolution 264, with an
improved star selection, based on the Hipparcos Input Catalogue (HIC) and an
empirical color relation between the visual magnitude (selectable in the HIC)
and the 15 m flux. All 83 stars observed presented a signal adequate
for an accurate centroid determination. Only stars were used for which
Hipparcos data on position and proper motion was available. Both the magnitude
and the orientation of the error vector were confirmed to follow the position
of the guide star on the STR.
Following the indications from the first experiment, an independent study was
undertaken by the Spacecraft Control Centre. At the start of each revolution a
``confirmation'' star is tracked simultaneously with the first guide star used,
in order to confirm the accuracy of ISO 's attitude. This information was used
to reconstruct the absolute pointings over a range of 120 revolutions. The APE
was found to be a strong function of the position of the guide star on the STR
and of the Solar Aspect Angle. A model was developed which involves a
non-uniformity of the STR focal length and fine tuning of various alignment
parameters, and was seen to predict down to < 1 accuracy the pointing errors
as measured by ISOCAM during the revolution 264 experiment.
Fine tuning a major satellite system beyond its specification means ripple down to all associated subsystems; as one of the s/c axis is commanded to point in the direction of the Sun, any uncertainty in the determination of the Sun position directly translates into a pointing uncertainty. A refined model of the Sun position was developed, which takes now into account factors previously deemed negligible, such as the ISO orbit parallax, the velocity of light and fine gridded ephemerides.
Both models (non-uniformity of STR field of view and Sun model) were verified
to improve the pointing accuracy to < 2 in another dedicated
experiment. This used the SWS in revolutions 368/369, with various pointings
around HR6705.
A further improvement, down to 1 , is expected soon, by (i) letting
the Sun position model on-board parameters be updated during a revolution and
(ii) post-processing of the SWS Focal Plane Geometry, based on the knowledge of
the measured beam profiles. It is also expected that appropriate
flux/wavelength corrections will in the near future be possible on the ISO
data, when an a-posteriori pointing reconstruction algorithm will be integrated
in the pipeline, based on the same model now used in the uplink.