The constant demand of LWAC applications in structural engineering increases the need of performing studies focused on the behavior of reinforced members produced with LWAC. Mainly, by combining reduced weight and good mechanical performance, LWAC is an efficient solution for flat slabs. The design methods of punching shear strength of LWAC slabs are commonly based on experimental studies of NWC. However, both the stress–strain relation and the distribution of internal stresses of LWAC are quite different from those of NWC, due to the stiffness compatibility between the binding matrix and the LWA, and due to the enhanced performance of interfacial transition zone LWA-matrix. This behavior influences the distribution of internal stresses of LWAC, when compared with NWC, and results in a linear stress–strain relation until around 90 % of maximum stress and a brittle failure after peak when unconfined (Costa in, Lightweight aggregate structural concrete: precast and strengthening of existing structures, 2012). This difference is ignored by the main structural concrete codes or the design expressions of NWC are modified by a corrective coefficient for LWAC, depending on its density. This paper presents an experimental study focused on punching capacity of LWAC slabs. Six slabs were produced with equal longitudinal reinforcement ratio and without shear reinforcement, varying the compressive strength of LWAC from 29 to 54 MPa. Based on the recorded data during the tests, the cracking and maximum loads, the displacements, rotations and stiffness, the failure modes and cracking patterns are presented and analyzed. Experimental results were compared with design predictions of main codes, namely, EC2, MC2010 and ACI318. The results revealed that the variation of LWAC strength influences the punching strength, but has no significant effect on the stiffness and on the angle of the main crack of punching cone. The evaluation of punching shear strength achieved by design methods is higher than the experimental results and, in the case of MC2010 with level I of approximation, is more than double. Excepting for EC2, the ratio between the maximum experimental punching strength and the corresponding code prediction decreases with the increase of LWAC strength.