i'm not sure if it makes sense to integrate all the darks together... i'd keep them separate and then just use the dark that's closest to the length of the lights you are calibrating.
That's an interesting point. I took this to also mean that the combination of darks may vary in exposure times. Combining a set of darks with a sum of 1000 seconds (by way of example) may involve combining different exposure lengths. Correct me if I'm wrong, but I don't think it matters. 
I presume that is the case for darks to lights also.
Don't worry if your master dark has 1000 seconds of exposure and your flat frame is only 10 seconds: IC will multiply the thermal noise of the master dark by 0.01.
i took that to mean that the dark subexposures are 1000s each, and that the master dark is composed of some unknown number of them. 1000s is only 16.67 minutes, which is more like the length of a single subexposure than the sum total of all the exposures in a stack.
Thanks for your comments guys.
I didn't mention but my lights are at different exposures to. I have
Lights 6x600secs, 10x300secs, 20x120 secs and 15x60secs
Darks 10x600secs, 10x300secs, 10x120secs and 10x60secs
Flats 20
No bias.
I want to calibrate and stack in Pix. Are you suggesting that I stack and calibrate each different exposure to end up with 4 images then integrate them together to get the final image? I feel the need for a tutorial 
Thanks for the continued help......I really need it.
okay, first off, you have to use a master bias frame if you expect to scale the master dark. the bias noise is fixed in every kind of frame, and if you don't remove it from the master dark, when you scale the master dark, you will also scale the bias noise. this will end up over or under-subtracting the bias noise from your lights.
also, it's hard to tell from a single line in a forum post, but you might have the flow backwards. first you calibrate each light frame with your calibration masters. then you register and stack the calibrated lights. but you should only stack together lights of the same duration, so yes, you should end up with 4 stacks. if you register each subexposure against the same reference image, then the stacks should already be aligned to each other, so you can then go right into HDRComposition and merge together your 4 exposure lengths. then carry on with normal processing.
It is my understanding that you can integrate all your darks and create a single master - same for bias and flats.
I probably conflict with pfile here, but as the master is a master, I think it is intended to be an integration of all the darks and through scaling is representative of all the lights (if that makes sense). Happy to recant if I'm wrong about this.
You can integrate individual time sets, with HDRComposition for instance if you want to reveal detail - e.g., M42.
the goal of integrating any image is to reduce the uncertainty of the value of a given pixel as much as possible. this goes for signal as well as 'noise' caused by dark current or bias current. if your subexposures have different lengths, then the value of a given pixel is going to have multiple distributions around different mean values. therefore there is no one true mean value for a given pixel to 'home in' on, and so your pixel value does not converge properly. i've always interpreted this to mean that your subexposures should have the same exposure time, whether they are darks or flats or lights. during calibration, PI is scaling the master dark to remove the dark noise from the light subexposure under consideration, to compensate for the master dark being at a different temperature or subexposure length than the light. but that's okay because if the master dark is 'clean' (meaning the uncertainty of the pixel values is low because you have used many dark subexposures to come up with the master dark), then you can come up with a linear scaling value for the dark before subtracting it from the light.
