> is there a good manual about the preprocessing procedures in spm? 

The manual is pretty comprehensive (256 pages), but we don't necessarily
have all the answers.  Everybody has different data and questions that
they want to attempt to answer with that data.  The choices you make in
your analysis will depend on what works best, and empirical exploration
is sometimes needed.  Chapters 25 and 26 of the manual describe
procedures that work well for data from the FIL.  I would suggest using
these procedures by default, but it is possible that a different
strategy may be better for your particular data.  Similarly, many of the
tools have default settings.  In most cases, I would suggest using these
values.

Just as there is no "canonical design matrix" for fitting any general
linear model through time series data, there is also no single
pre-processing strategy that works best for everything.  There are many
different imaging modalities, field strengths, sequences etc.  Each of
these has its own artifacts, image intensity distributions etc.  Repeat
times, fields of view etc are all very variable.  Some people collect
field maps for correcting distortions.  Some scans have more or less
distortion or dropout than others.  All these things should influence
the choices that are made.

> - do i get better results in normalisation with coreg my epi on the t1

> image and normalize the latter with the t1 template (there is a mail
> in the spm list archives, i think it was john ashburners, who prefers
> normalization with the mean epi image, but it was according spm99)? 

This was applicable six years ago, but hopefully the software (and
data??) are now rather better than they were back then.  This strategy
worked reasonably well for data from the FIL scanner, because the tissue
intensities and artifacts in the data were similar to those in the
templates.  Back in the late 20th century, spatial normalization was
done by warping the subject's images so that the intensity differences
between the image and template were minimized.  The relative intensities
of the different tissue types had to be similar for the template and
individual image for this to work.

Things can be done slightly differently by SPM5.  If you use the Segment
button for estimating spatial normalization parameters, then no
assumptions about the intensity of the different tissue types are made -
other than grey matter is all about the same intensity, and that white
matter and CSF also have fairly uniform intensities.  The Segment
procedure also models out intensity non-uniformity artifact.  In the
past, many of the spatial normalization procedures encountered problems
because of this artifact - but the situation should be slightly improved
by now.

It is still possible that using the Segment button with your EPI data
could produce more accurate spatial normalization than estimating the
warps from your anatomical images, and applying them to your
coregistered fMRI data.  This will mostly depend on the amount of
distortion in your fMRI.  Because coregistration only estimates a
rigid-body transformation, then it may not necessarily determine an
accurate alignment between the functional and anatomical data.  Even if
your spatially normalized anatomical data are all in alignment with MNI
space, if the anatomical and functional data are not in register, then
the spatially normalized functional data may not be so accurately
aligned with each other.  This is something that is data dependant, and
may need empirical exploration in order to figure out the best strategy.

It is possible that distortions in fMRI data have not been taken
seriously enough by us SPM authors.   In fact, I'm not sure they are
taken seriously enough by others either - especially in terms of the
effects of such mis-registration on surface based analysis methods.


> or - do i have to estimate & resclice images in the coreg step or is
it
> enough to just estimate (by the way, reslicing "blows" up my single
epi
> images to a size of the t1 image (160mb) and therefore it takes
hours). 

As Simon points out, you don't need to do any reslicing at this stage
(maybe I'll take out the Coregister and Reslice option from the next SPM
release).  Avoiding such un-necessary reslicing was my main motivation
for introducing the .mat files that confused (and still confuse) so many
people for so long.  These .mat files are no longer used (except for 4D
data).  Image orientations are now written straight into the .hdr files.
The Check reg button is useful for seeing if your images are considered
by SPM to be in register with each other.


> the spm5 manual is no real help, because it just describes the single 
> features, but gives no advise what do do. the typical related papers
> are no real help either, because their are overloaded with
mathematical
> terms and procedures, i do not understand.
> so anything around for a dummy who just want to drive a car without
> being a mechanic?

Currently, SPM can be thought of as a bunch of components - rather than
a complete car, so you'll sometimes have to be a mechanic as well as a
driver.  Making it work in good driving conditions is generally fairly
straightforward, but customization is sometimes needed for other
situations.  One of the features of SPM is that it is relatively easy to
get at what is under the bonnet.

Best regards,
-John