Comments on Global Parton Analyses

We discuss the causes which can limit the accuracy of the predictions based on the conventional PDFs when including in global parton analyses the data at moderate scales $\mu$. The first is the existence of power corrections ${\cal O} (Q_0^2/\mu^2)$ due to the double counting of contributions arising from the region below the input scale $Q_0$. The second concerns the possible inclusion of the BFKL re-summation of the $(\alpha_s\ln (1/x))^n$ terms. The third is the treatment of the heavy-quark thresholds. We show how to include the heavy-quark masses ($m_h$ with $h=c,b,t$) in DGLAP evolution which provides the correct {\it smooth} behaviour through the threshold regions and how to subtract the low parton virtuality $|k^2|<Q^2_0$ contributions from the DIS and Drell-Yan NLO coefficient functions in order to avoid the double counting.