Basically it is a specification that has been determined by past
testing and documentation of the given fastener type plus engineering
consideration of the assembly parts in question.  The two most
important things regarding the fastener are 1) it must not break/strip
threads at this torque value and 2) it must not back out from lack of
tension.  The makeup of the parts and their intended duty modify this
equation as well.

In the case of a final assembly, lots of data is reviewed before
assigning a value to a fastener.  Machining and material spec data
publications are used by design engineers as a baseline towards
engineering their specific requirements.

Toyota MDT in MO

They are determined by "tensile strength"

Here's an interesting link for ya:
http://www.raskcycle.com/techtip/webdoc14.html

In a nutshell.... Bolted connections work because tightening the bolt
generates a clamping force between the two items. The trick is to
generated as much clamping force as possible without damaging the bolt
by permanently stretching it. In a range, a bolt can be
tightened(stretched) and loosened without permanent change in length-
it will return to its original length. If you stretch it too far, it
"yields" and will not return to its original length. The best amount of
tightening would be to about 90% of the point where it wouldn't return
to its original length-- a little safety factor for the unknown. Now
the trick is to try to relate "torque" to the stretch in the bolt,
including all kinds of other factors like type of metal, strength,
thread pitch, lubrication, etc. Testing and usage has verified common
engineering formulas for this and people have set up tables for common
bolt size torque values. If you think about it, "threads per inch" tell
you that with one turn of the bolt, exactly how far it will stretch.
Say... if you have 16 threads per inch, then one turn will stretch the
bolt 1/16" of an inch.

This is a very complicated issue. In general the engineers would
determine a target clamping force and work backwards from there to
determine a tightening torque. For a gasketed joint, they would
determine a bolt pattern and pick a bolt size to provide the required
clamping force (and consider the mechanical strength for a load
carrying member). The bolts would be sized so that when properly
torqued they would be at something like 80% of their proof load. The
torque needed to achieve this clamping force is a function of the
materials involved (bolt material and base material for female
threads), plating (these affect the frictional characteristics of the
joint as the bolt is torqued), lubrication (another contributor to
frictional characteristics), head style, etc. The type of loads the
bolt would be subjected to would also be important (axial, shear,
combination, static, varying, etc.) For a joints subjected to varying
axial loads, they would go for a higher percentage of the proof load
(90 to 100%). Most modern head, rod, and main bearing cap bolts are
torqued to the yield point. This is done using specialized machines
that sense a change in the stress strain relationship as the bolt
reaches the yield point. This is a very good way to torque bolts for
variable loads, but difficult to duplicate in a repair environment.

I have some information of torque settings at
http://home.mindspring.com/~ed_white/id9.html , but it is simple
information related to steel on steel joints with non-reversed
stresses.

Ed