John Payson summarised it well.  PWM works for pure resistive loads.
It is not appropriate for loads with a back-EMF.  For this, you need to
use an inductor and commutation diode to convert the PWM to a current
drive.

  This however, appears at first glance to prohibit feedback using back-
EMF.  Not so.  The model given by John was the winding resistance in
series with the back-EMF, so if you know the voltage across the motor
and the current, you can deduce the voltage drop across the winding
resistance and by subtraction, the back EMF while it is powered.

  In fact, this method even works on series-wound motors, as you measure
whilst the field windings are energised (and if you want the speed more
precisely, you divide the imputed back-EMF by the current which implies
the field strength).
--
  Cheers,
        Paul B.