In recent years, USB flash drives have become very popular.  They can
plug into almost any PC and they're automatically recognized.  Why not
try to interface one to a microcontroller based project, so it can the
USB flash drive's memory?

The short answer: USB is a complex and difficult protocol to implement.

Compact Flash (used in most digial cameras) is much better suited for
interface to a microcontroller.  However, for hobbists and students who
already own and use a USB flash drive, this advise is rarely appealing.
Switching to Compact Flash and buying a USB-based reader for CF just
doesn't seem the same as directly supporting a USB flash drive.

Having said USB is complex and difficult, it's very easy on high-end
systems (like PCs) which include all the hardware and software needed
to communicate with USB devices, because the chipset vendor and operating
system programmers have done a lot of work.  To implement USB on some
other system that doesn't already have all the work done, the first
source of information is the USB specification.  Fortunately, it is a
free download from www.usb.org

http://www.usb.org/developers/docs/usb_20.zip

The main PDF file is 650 pages.  However, chapters 3, 4 and 5
(titled "Background", "Architectural overview" and "USB Data Flow
Model") span only about 70 pages with a very good overview of the
most important concepts.

After reading those 70 pages, you should have a very clear that
USB has only one "host", possibly many "devices" and maybe some
"hubs" in between them.  The important point is that USB is not a
peer to peer protocol.  In USB, there is only 1 "host" which is a
much more complex design that is responsible for controlling all
communication on the bus.  The "devices" are much simpler and only
need to respond to queries from the "host".

USB flash drives are USB devices, so to communicate with one, you
will need to implement the complex host functionality.  Once you
can control the bus and query the endpoints, you will then need the
USB Mass Storage specification to learn how to read and write sectors.
The good news is that only "control" and "bulk" endpoints are used
(the USB Mass Storage 1.1 spec, which uses "interrupt" transfers,
was never adopted by Microsoft and as a result almost all USB key
drives only really support the 1.0 bulk-only protocol).  Once you
can read sectors from the drive, you will need to learn about the
FAT16 filesystem.  A difficult-to-understand specification is
available from Microsoft.  I have written a page that attempts to
explain FAT32 (very similar) for beginners:

http://www.pjrc.com/tech/8051/ide/fat32.html

If you managed to read the 70 pages in chapters 3, 4 and 5 of the
USB specification, you should understand that implementing the USB
host functionality is a large and complex project.  In the mid-90s,
USB was often called the "useless serial bus", because it took
Microsoft YEARS to finally include a quality USB implementation in
the latter OSR2 versions of windows 95 and finally windows 98.

Implementing USB host functionality is not a simple project to be
taken lightly, and certainly not something most students could be
expected to accomplish within the scope of a 1 year project.  If you
must interface to a USB flash drive or other USB device, seek out a
chip that implements USB host features and comes with the OHCI or
EHCI code that implements the complex management of the USB bus.

Or simply switch to Compact Flash, which is fairly easy to interface
to most microcontrollers.