Action potentials have a central role in the nervous system and in many cellular processes, notably those involving ion channels. The accurate measurement of action potentials requires efficient coupling between the cell membrane and the measuring electrodes. Intracellular recording methods such as patch clamping involve measuring the voltage or current across the cell membrane by accessing the cell interior with an electrode, allowing both the amplitude and shape of the action potentials to be recorded faithfully with high signal-to-noise ratios. However, the invasive nature of intracellular methods usually limits the recording time to a few hours, and their complexity makes it difficult to simultaneously record more than a few cells. Extracellular recording methods, such as multielectrode arrays and multitransistor arrays, are non-invasive and allow long-term and multiplexed measurements. However, extracellular recording sacrifices the one-to-one correspondence between the cells and electrodes, and also suffers from significantly reduced signal strength and quality. Extracellular techniques are not, therefore, able to record action potentials with the accuracy needed to explore the properties of ion channels. As a result, the pharmacological screening of ion-channel drugs is usually performed by low-throughput intracellular recording methods. The use of nanowire transistors, nanotube-coupled transistors and micro gold-spine and related electrodes can significantly improve the signal strength of recorded action potentials. Here, we show that vertical nanopillar electrodes can record both the extracellular and intracellular action potentials of cultured cardiomyocytes over a long period of time with excellent signal strength and quality. Moreover, it is possible to repeatedly switch between extracellular and intracellular recording by nanoscale electroporation and resealing processes. Furthermore, vertical nanopillar electrodes can detect subtle changes in action potentials induced by drugs that target ion channels.