Arne, Brian, Dan, Lew, Michael, Taichi, Many thanks for the comments. These have pointed out a direction to proceed. Firstly, I'll get rid of the short exposures as much as possible by reducing the aperture size of the telescope with a mask. A diameter of about 120mm (with the 80mm secondary) will increase the times by 4x to 6x. This should help the vignetting also. I have always been reluctant to go below 10 seconds integration and at Auckland we'd always collect 3 or 6 of these. Longer exposures will help with the atmospheric effects but should also get rid of the shutter imprecision. For very bright objects I'll just do more exposures and hope these reduce the problem. Secondly, I'll also use separate flat fields in each colour. I didn't think of the dust on the filters and hoped that one would do. It'll mean some changes to the reductions but this is simple. Oddly enough, the R and I measures were the most consistent. The software apertures used are about 30 arcseconds. I don't do PSF, the focal length is too short at 50 inches and the CCD resolution is quite coarse. I set the inner four apertures in Munidos and check the progression of magnitudes to ensure that almost all the signal is in the innermost one. Usually this is 1-3% fainter than the fourth aperture. In the low-res mode the ST6 reduces the field by adding pixels with a 2:1 ratio in one direction, a 3:2 ratio in the other. I'm not too sure how the 3:2 ratio is achieved but it may create an elliptical aperture - which is probably unimportant. I note Michael's comments about out of focus images. Mike Bessell mentioned that there should be a 2x2 or 3x3 pixel coverage for each star to reduce pixel to pixel variations. So we tend to defocus the images to achieve this. But in the low-res mode some extra defocussing might be useful. I'll try a little more. Answering Arne's questions about checking the shutter speed - I could get someone to try to do this but it seems easier to just stop down the telescope and make this factor less important. I'll worry about this if the accuracy doesn't improve. I appreciate that the sky conditions affect this as well, so I was hoping that someone could make some comments based upon experience with very short exposures of 0.1 to 5 seconds. The stars are all bright so there are plenty of photons. Since the low-res mode bins the pixels by adding, the operating region is well below the pixel saturation level. At most it can be 33%. And Ive been setting the maximum at about 2/3 of this level. Checking by examining bright images in CCDOPS as they are acquired. Differential photometry? You could almost call it this in the old pm tube sense in that the observations are all made within 30 degrees of the zenith and that a couple of comparison sets are taken every 30 minutes. But not in the CCD sense in that the comparison stars are NOT on the same image. In the magnitude 5 to 9 range there are not often suitable stars available. I have two in 30 fields. I use the Cape E region standards at S45 for the comparisons - the CCD won't go through the forks with the filter system (it is the latest model CFW-8) so can't get at the F regions or the LMC SMC sequences. I mentioned the location near the sea as I think this may induce low level transparency variations. The people down near Wellington thought they were getting these. In Auckland at a city site the few photometric nights (10% to 15% perhaps) were usually very steady. Here I get far more nights (not in 1999 so far) but maybe the transparency variations are greater. But, once again, the variations mentioned in my first enquiry were unexpectedly large over a short time scale ( up to 5% over 1 to 5 minutes). Your replies will enable me to get rid of all the instrumental errors so that the remainder will have to be atmospheric. At that stage I'll see whether the measures are good enough to do what was intended - determine annual epochs for Cepheids to look for period changes (and SR star to a lesser degree, where the measures are quite good enough). Even at 2% accuracy 10-15 measures a season should be good enough. This works out at about an hour or two a star, which is acceptable. Regards and thanks to all - it really illustrates the value of VSNet Chat, Stan