In case you don't manage to get to the info that Usasma suggested, a summary:
Failure to create a dump generally happens for one of the following reasons:
1) The crash wasn't actually a "bugcheck". For example, spontaneous reboots or power-down events caused by hardware anomalies do not cause KeBugCheckEx (the "bluescreen function") to be called. If you're actually seeing the white-on-blue text and "dumping physical memory..." message, this obviously isn't what's happening in your case.
2) The pagefile on the boot partition (where \Windows lives) is insufficiently large or it doesn't exist at all. That's actually the only pagefile that matters for the purposes of dump creation. Irrespective of the dump file that's configured via sysdm.cpl, the dump is initially created by writing pages from physical memory into the boot partition pagefile, which is then renamed to "memory.dmp" and moved on reboot as configured (another pagefile is created).
The size of the necessary pagefile depends on whether the dump type is set to "kernel" or "complete". Kernel is sufficient in the vast majority of cases, and it's all you need in this instance. For a kernel dump, the pagefile on the boot partition must be large enough to accomodate the resident portions of the kernel-mode virtual address range, and that'll vary from say 200MB on lower-end machines to 2GB on monster x64 boxes. Unless your machine is a busy server, a 2GB pagefile on the boot partition is perfectly adequate for the purposes of dump file generation.
Minidumps are actually generated from kernel dumps, so if you specify a "small" dump file (minidump), all that happens is the system dumps a kernel file, summarises it into a minidump, and then doesn't bother with the "memory.dmp".
3) The most complex dump generation problems are those which take out the OS's ability to write data in the first place. For example, crashes very low in the disk and volume management drivers can sometimes prevent the committment of memory pages to disk. The only way to deterministically work out what may be happening is to run the machine under a kernel debugger, and that is a 200-level exercise. The book Usasma mentioned has details on the procedure, besides being one of the best books on Windows.
Good luck with it.
