A retrospective analysis of the dose delivery system (DDS) performances of the initial clinical operation at CNAO (Centro Nazionale di Adroterapia Oncologica) is reported, and compared with the dose delivery accuracy following the implementation of a position feedback control. Log files and raw data of the DDS were analyzed for every field of patients treated with protons and carbon ions between January 2012 and April 2013 (~3800 fields). To investigate the DDS accuracy, the spot positions and the number of particles per spot measured by the DDS and prescribed by the treatment planning system were compared for each field. The impact of deviations on dose distributions was studied by comparing, through the gamma-index method, 2 three-dimensional (3D) physical dose maps (one for prescribed, one for measured data), generated by a validated dose computation software. The maximum gamma and the percentage of points with gamma≤1 (passing volume) were studied as a function of the treatment day, and correlated with the deviations from the prescription in the measured number of particles and spot positions. Finally, delivered dose distributions of same treatment plans were compared before and after the implementation of a feedback algorithm for the correction of small position deviations, to study the effect on the delivery quality. A double comparison of prescribed and measured 3D maps, before and after feedback implementation, is reported and studied for a representative treatment delivered in 2012, redelivered on a polymethyl methacrylate (PMMA) block in 2018. Systematic deviations of spot positions, mainly due to beam lateral offsets, were always found within 1.5mm, with the exception of the initial clinical period. The number of particles was very stable, as possible deviations are exclusively related to the quantization error in the conversion from monitor counts to particles. For the chosen representative patient treatment, the gamma-index evaluation of prescribed and measured dose maps, before and after feedback implementation, showed a higher variability of maximum gamma for the 2012 irradiation, with respect to the reirradiation of 2018. However, the 2012 passing volume is >99.8% for the sum of all fields, which is comparable to the value of 2018, with the exception of one day with 98.2% passing volume, probably related to an instability of the accelerating system. A detailed retrospective analysis of the DDS performances in the initial period of CNAO clinical activity is reported. The spot position deviations are referable to beam lateral offset fluctuations, while almost no deviation was found in the number of particles. The impact of deviations on dose distributions showed that the position feedback implementation and the increased beam control capability acquired after the first years of clinical experience led to an evident improvement in the DDS stability, evaluated in terms of gamma-index as a measure of the impact on dose distributions. However, the clinical effect of the maximum gamma variability found in the 2012 representative irradiation is mitigated by averaging along the number of fractions, and the high percentage of passing volumes confirmed the accuracy of the delivery even before the feedback implementation.