We propose a novel attention model based on gated recurrence, which we call gated recurrent attention (GRA). GRA controls the contextual information by employing two gates. To show the GRA's alignment and style modeling performance, we upload some samples synthesized by Tacotron-GST with the location-sensitive attention (LSA) and GRA. The models were trained only on MAILABS-US corpus then synthesized the samples with style-references in the MAILABS-US and VCTK.