<p>Recently, deep convolutional neural networks have made outstanding progress on single image super-resolution (SISR). However, most of these methods improve the performance by increasing the model size, which hinders their applications on resource-constrained scenarios. To address this issue, we propose a novel attention interaction and recalibration network (AIRN) for lightweight SISR, where the expressive features extracted by attention blocks are recalibrated and interacted with other blocks. AIRN consists of cascaded attention interaction-recalibration blocks (AIRBs). An AIRB is constructed with several spatial attention and channel attention refining blocks (SCARBs) via features reused connections. The information extracted by spatial attention and channel attention is recalibrated in a SCARB and interacted between SCARBs, guiding the network to excavate more expressive features. This interaction and recalibration strategy has improved the ability of feature presentations, thereby reducing the number of parameters. Despite having only approximately 300K parameters, our AIRN demonstrates superior SISR results on benchmark datasets, compared to other advanced models at similar parameter level. In most datasets, AIRN surpasses competitors in peak signal to noise ratio by over 0.1 dB.</p>