We propose a new method to detect the gravitational lenses in the ongo
ing microlensing experiments using medium and high resolution spectros
copy (lambda/Delta lambda > 6000). Since the radial velocity of the le
ns and lensed source typically differs by similar to 100 km s(-1), the
spectral lines from the lens and source will be shifted relative to e
ach other by (1 - 2) Angstrom in the optical. We simulate realistic co
mposite spectra assuming different spectral types for the lens and sou
rce and study the lens detectability as a function of the signal-to-no
ise ratio, spectral resolution and lens-to-source light ratio. We show
that it is possible to measure the difference in radial velocity from
an unequivocal signature in the difference of cross- and auto-correla
tion functions calculated from two spectra obtained at different magni
fications. If the lens is brighter than 10% (Delta m(v) similar to 2.5
) of the unmagnified source we find that a spectral resolution of lamb
da/Delta lambda similar to 6000 and a signal-to-noise of 50 (at magnif
ication maximum) are sufficient to determine the relative radial veloc
ity of the lens. At lambda/Delta lambda = 40000, the spectral resoluti
on of high resolution spectrographs of 8-10 m class telescopes, the le
ns could even be detected at a brightness of similar to 3% (nm, simila
r to 4.0) of the source. Radial velocities higher than 50 km s(-1) can
be measured with an accuracy of a few km s(-1). Practical difficultie
s and observation strategies are also discussed.