This paper presents the results of a detailed X-ray spectral analysis of a sample of 123 X-ray sources detected with XMM-Newton in the Lockman Hole field. This is the deepest observation carried out with XMM-Newton with more that 600 ks of good EPIC-pn data. We have spectra with good signal to noise (>500 source counts) for all objects down to 0.2−12 keV fluxes of ∼5 × 10 −15 erg cm −2 s −1 (flux limit of ∼6 × 10 −16 erg cm −2 s −1 in the 0.5−2 and 2−10 keV bands). At the time of the analysis, we had optical spectroscopic identifications for 60% of the sources, 46 being optical type-1 AGN and 28 optical type-2 AGN. Using a single power law model our sources' average spectral slope hardens at faint 0.5−2 keV fluxes but not at faint 2−10 keV fluxes. We have been able to explain this effect in terms of an increase in X-ray absorption at faint fluxes. We did not find in our data any evidence for the existence of a population of faint intrinsically harder sources. The average spectral slope of our sources is ∼1.9, with an intrinsic dispersion of ∼0.28. We detected X-ray absorption (F-test significance ≥95%) in 37% of the sources, ∼10% in type-1 AGN (rest-frame N H ∼ 1.6 × 10 21 −1.2 × 10 22 cm −2 ) and ∼77% (rest-frame N H ∼ 1.5 × 10 21 −4 × 10 23 cm −2 ) in type-2 AGN. Using X-ray fluxes corrected for absorption, the fraction of absorbed objects and the absorbing column density distribution did not vary with X-ray flux. Our type-1 and type-2 AGN do not appear to have different continuum shapes, but the distribution of intrinsic (rest-frame) absorbing column densities is different among both classes. A significant fraction of our type-2 AGN (5 out of 28) were found to display no substantial absorption (N H < 10 21 cm −2 ). We discuss possible interpretations to this in terms of Compton-thick AGN and intrinsic Broad Line Region properties. An emission line compatible with Fe Kα was detected in 8 sources (1 type-1 AGN, 5 type-2 AGN and 2 unidentified) with rest frame equivalent widths 120−1000 eV. However weak broad components can be easily missed in other sources by the relatively noisy data. The AGN continuum or intrinsic absorption did not depend on X-ray luminosity and/or redshift. Soft excess emission was detected in 18 objects, but only in 9 (including 4 type-1 AGN and 4 type-2 AGN) could we fit this spectral component with a black body model. The measured 0.5−2 keV luminosities of the fitted black body were not significantly different in type-1 and type-2 AGN, although the temperatures of the black body were slightly higher in type-2 AGN ( kT = 0.26 ± 0.08) than in type-1 AGN ( kT = 0.09 ± 0.01). For 9 sources (including 1 type-1 AGN and 3 type-2 AGN) a scattering model provided a better fit of the soft excess emission. We found that the integrated contribution from our sources to the X-ray background in the 2−7 keV band is softer (Γ = 1.5−1.6) than the background itself, implying that fainter sources need to be more absorbed.