We present an implementation of the difference image photometry based on the Alard and Lupton optimal PSF matching algorithm. The most important feature distinguishing this method from the ones using Fourier divisions is that equations are solved in real space and the knowledge of each PSF is not required for determination of the convolution kernel. We evaluate the method and software on 380G B of OGLE-II bulge microlensing data obtained in 1997-1999 observing seasons. The error distribution is Gaussian to better than 99% with the amplitude only 17% above the photon noise limit for faint stars. Over the entire range of the observed magnitudes the resulting scatter is improved by a factor of 2-3 compared to DoPhot photometry, currently a standard tool for massive stellar photometry in microlensing searches. For testing purposes the photometry of approx 4600 candidate variable stars and sample difference image data are provided for BUL_SC1 field. In the candidate selection process, very few assumptions have been made about the specific types of flux variations, which makes this data set well suited for general variability studies, including the development of the classification schemes.