Procedures based on current methods to detect sources in X-ray images are a
pplied to simulated XMM-Newton images. All significant instrumental effects
are taken into account, and two kinds of sources are considered - unresolv
ed sources represented by the telescope PSF and extended ones represented b
y a beta -profile model. Different sets of test cases with controlled and r
ealistic input configurations are constructed in order to analyze the influ
ence of confusion on the source analysis and also to choose the best method
s and strategies to resolve the difficulties. In the general case of point-
like and extended objects the mixed approach of multiresolution (wavelet) f
iltering and subsequent detection by SExtractor gives the best results. In
ideal cases of isolated sources, flux errors are within 15-20%. The maximum
likelihood technique outperforms the others for point-like sources when th
e PSF model used in the fit is the same as in the images. However, the numb
er of spurious detections is quite large. The classification using the half
-light radius and SExtractor stellarity index is successful in more than 98
% of the cases. This suggests that average luminosity clusters: of galaxies
(L[2-10]keV similar to 3 10(44) erg/s) can be detected at redshifts greate
r than 1.5 for moderate exposure times in the energy band below 5 keV, prov
ided that there is no confusion or blending by nearby sources. We find also
that with the best current available packages, confusion and completeness
problems start to appear at fluxes around 6 10(-16) erg/s/cm(2) in [0.5-2]
keV band for XMM-Newton deep surveys.