It is demonstrated that the fragmentation functions at large momentum fraction play a key role in hard hadron production from relativistic proton-proton collisions. We find that this region of the fragmentation functions is not strongly constrained by the electron-positron data. This freedom can be used (together with the transverse momentum distribution of partons) to reproduce hard pion-to-proton ratio data in relativistic proton-proton collisions.PACS 13.87.Fh Hadron spectra at large transverse momentum (p T ), where perturbative Quantum Chromodynamics (pQCD) has good predictive power, are very important for our understanding of the physics at the Relativistic Heavy Ion Collider (RHIC) and at the planned Large Hadron Collider (LHC). Various experiments at these colliders focus on hard (high p T ) hadron spectra. Jet "tomography" (the study of the strongly-interacting medium via energy loss of hard partons) has been proposed to detect the quark-gluon plasma (QGP), using high-p T hadron production [1,2].Suppression of total charged hadron production in Au + Au collisions relative to a nucleon-nucleon reference is reported at RHIC [3,4]. However, for the proton-to-pion (p/π) ratio the PHENIX experiment reports an anomalous enhancement in Au+Au collisions at √ s = 130 GeV [5]. An explanation for the p/π enhancement was proposed recently, combining pQCD with soft physics and jet quenching [6]. It should be kept in mind in this context that, while pQCD is quite successful for total charged hadron (h + + h − ) and pion production at large p T in pp collisions, proton production in pp is not well understood using the language of pQCD. In fact, pQCD underestimates the p/π + ratio by a factor of 3-10 in pp collisions (see Fig. 3(a) of this work for √ s = 27.4 GeV, and e.g. Ref.[6] for Tevatron energy).In this Letter we look into how pQCD is used to calculate p T spectra in pp collisions. We focus on the role of a non-perturbative ingredient, the fragmentation function (obtained by fitting data) in the production of hadronic final states. Most of the information included in the fits comes from h + + h − data. The fragmentation function (FF) of pions is studied in some detail. Less direct information is available on kaons, and the FF of protons is even less well-known. In the following, we concentrate on the proton FF as an example of the role of the FF in the hadroproduction of hard particles. We find (for all types of hadrons) that the value of the FF in a region of phase space where it is least constrained plays a decisive role for hadron production at RHIC and LHC.To predict the p T spectra of final state hadrons in the framework of pQCD, perturbative partonic cross sections need to be convoluted with parton distribution functions (PDF-s) and fragmentation functions (FF-s) according to the factorization theorem [7]. Perturbative QCD has nothing to say about the details of FF-s, apart from describing their scale evolution. The FF-s are assumed to be universal, meaning that once extracted from a limited set of data...