We present a new metric to characterize Transactional Memories (TMs): the input acceptance. This metric represents, for a given TM, its ability to commit transactions depending on the  interleaving (i.e., schedule) of their actions. Up to now, the main evaluation metric was the number of committed transactions by time unit (a.k.a. throughput), however, throughput does not capture the likeliness for a TM to commit a transaction. 

Unlike throughput, the input acceptance of a TM indicates the quantity of given schedules for which the TM commits its transactions.  The difficulty in designing a correct TM comes more from ensuring that some serializable transactions commit than from ensuring that non-serializable transactions abort. 

We identify few TM designs shared by several existing TMs and we compare them along with this new metric. Our theoretical results, confirmed by experimental results, totally order the input acceptance of these designs.