We discuss the prospect of deblending microlensing events by observing astrometric shifts of the lensed stars .
Since microlensing searches are generally performed in very crowded fields , it is expected that stars will be confusion limited rather than limited by photon statistics .
By performing simulations of events in crowded fields , we find that if we assume a dark lens and that the lensed star obeys a power law luminosity function , n ( L ) \propto L ^ { - \beta } , over half the simulated events show a measurable astrometric shift .
Our simulations included 20000 stars in a 256 \times 256 Nyquist sampled CCD frame .
For \beta = 2 , we found that 58 \% of the events were significantly blended ( F _ { \ast } / F _ { tot } \leq 0.9 ) , and of those , 73 \% had a large astrometric shift ( \geq 0.5 { pixels } ) .
Likewise , for \beta = 3 , we found that 85 \% of the events were significantly blended , and that 85 \% of those had large shifts .
Moreover , the shift is weakly correlated to the degree of blending , suggesting that it may be possible not only to detect the existence of a blend , but also to deblend events statistically using shift information .