Electroencephalography　(EEG)　contains　rich　information　about　brain　activity.　Classification　based　on　motor　imagery　EEG　(MI-EEG)　is　essential　for　active　intelligent　rehabilitation　using　brain-computer　interface　technology.　However,　most　of　the　current　MI-EEG　decoding　methods　are　dedicated　to　the　study　of　efficient　time-frequency　feature　extraction　from　the　raw　EEG,　ignoring　the　spatial　features　associated　with　electrode　distribution.　Therefore,　this　paper　proposed　an　MI-EEG　classification　method　using　a　combination　of　three　dimensional　(3D)　spatial　interpolation　and　3D　convolutional　neural　network　(3DCNN)　to　fully　utilize　the　3D　spatial　features　of　electrodes.　First,　the　frequency　transformation　was　applied　to　the　EEG　signal　at　each　electrode　using　the　fast　Fourier　transform　and　the　energy　value　was　obtained.　Then,　the　3D　coordinates　of　the　scalp　electrodes　were　projected　into　the　3D　space　and　the　energy　values　were　interpolated　using　the　3D　interpolation　method　to　generate　a　3D　feature　image　containing　the　3D　real　spatial　location　information　of　the　electrodes.　Finally,　a　3DCNN　was　designed　to　match　the　3D　feature　image　for　feature　extraction　and　recognition.　The　proposed　method　obtained　77.19%　accuracy　in　the　BCI　Competition　IV　2a　dataset,　which　is　higher　than　the　existing　decoding　methods.　Results　from　experiments　validate　the　effectiveness　of　the　proposed　MI-EEG　classification　method.