The prevention of seismic risk at urban scale can be pursued through the estimate of the probability to reach or exceed a certain damage grade given the seismic input. In this framework, seismic fragility curves are nowadays of large interest as they express this probability in a synthetic way, also extended to large-scale applications. Real damage data are crucial in improving the reliability of the estimation of damage occurrence, although they can be influenced by a proper definition of the structural types and the completeness of observations. The paper shows the empirical fragility curves obtained for a sample of about 2250 masonry buildings located within 19 historical centers struck by the 2016 Central Italy earthquake. The damage grade was evaluated according to the European Macroseismic Scale (EMS-98), also considering undamaged buildings, at the end of the sequence that spanned between August and October. The buildings largely underwent several repairs and strengthening actions with reinforced concrete elements starting from the 1980s. The systematization of the structural features led to a taxonomy for strengthened and original buildings, which, basing on the observed damage patterns, was matched to an EMS-98 vulnerability class. The sample ranges from class A (worst behavior) to D (best behavior). Class A was typically given to original buildings (without interventions) or ill-advisedly tampered ones, i.e., those in which interventions had an unfavorable contribution to their seismic behavior. Class D described buildings with properly designed strengthening interventions, classes B and C intermediate situations. Fragility curves were obtained per each vulnerability class, as a function of the highest peak ground acceleration (PGA) observed in the sequence from shakemaps. The results were then compared to other empirical fragility models.