diff --git a/README.md b/README.md
index 2837b3e..803f29a 100644
--- a/README.md
+++ b/README.md
@@ -82,8 +82,6 @@ optional arguments:
                         Kernel size of causal convolution
   --hidden-ksize HIDDEN_KSIZE
                         Kernel size of hidden layers convolutions
-  --data-channels DATA_CHANNELS
-                        Number of data channels
   --color-levels COLOR_LEVELS
                         Number of levels to quantisize value of each channel
                         of each pixel into
@@ -127,8 +125,6 @@ optional arguments:
                         Kernel size of causal convolution
   --hidden-ksize HIDDEN_KSIZE
                         Kernel size of hidden layers convolutions
-  --data-channels DATA_CHANNELS
-                        Number of data channels
   --color-levels COLOR_LEVELS
                         Number of levels to quantisize value of each channel
                         of each pixel into
diff --git a/pixelcnn/model.py b/pixelcnn/model.py
index 1ff8eda..c41e42a 100644
--- a/pixelcnn/model.py
+++ b/pixelcnn/model.py
@@ -120,16 +120,18 @@ class PixelCNN(nn.Module):
     def __init__(self, cfg):
         super(PixelCNN, self).__init__()
 
+        DATA_CHANNELS = 3
+
         self.hidden_fmaps = cfg.hidden_fmaps
         self.color_levels = cfg.color_levels
 
-        self.causal_conv = CausalBlock(cfg.data_channels,
+        self.causal_conv = CausalBlock(DATA_CHANNELS,
                                        cfg.hidden_fmaps,
                                        cfg.causal_ksize,
-                                       data_channels=cfg.data_channels)
+                                       data_channels=DATA_CHANNELS)
 
         self.hidden_conv = nn.Sequential(
-            *[GatedBlock(cfg.hidden_fmaps, cfg.hidden_fmaps, cfg.hidden_ksize, cfg.data_channels) for _ in range(cfg.hidden_layers)]
+            *[GatedBlock(cfg.hidden_fmaps, cfg.hidden_fmaps, cfg.hidden_ksize, DATA_CHANNELS) for _ in range(cfg.hidden_layers)]
         )
 
         self.label_embedding = nn.Embedding(10, self.hidden_fmaps)
@@ -138,13 +140,13 @@ def __init__(self, cfg):
                                             cfg.out_hidden_fmaps,
                                             (1, 1),
                                             mask_type='B',
-                                            data_channels=cfg.data_channels)
+                                            data_channels=DATA_CHANNELS)
 
         self.out_conv = MaskedConv2d(cfg.out_hidden_fmaps,
-                                     cfg.data_channels * cfg.color_levels,
+                                     DATA_CHANNELS * cfg.color_levels,
                                      (1, 1),
                                      mask_type='B',
-                                     data_channels=cfg.data_channels)
+                                     data_channels=DATA_CHANNELS)
 
     def forward(self, image, label):
         count, data_channels, height, width = image.size()
diff --git a/sample.py b/sample.py
index d3425dd..08a45d6 100644
--- a/sample.py
+++ b/sample.py
@@ -16,8 +16,6 @@ def main():
     parser.add_argument('--hidden-ksize', type=int, default=7,
                         help='Kernel size of hidden layers convolutions')
 
-    parser.add_argument('--data-channels', type=int, default=3,
-                        help='Number of data channels')
     parser.add_argument('--color-levels', type=int, default=2,
                         help='Number of levels to quantisize value of each channel of each pixel into')
 
diff --git a/train.py b/train.py
index 763856b..7205647 100644
--- a/train.py
+++ b/train.py
@@ -87,8 +87,6 @@ def main():
     parser.add_argument('--hidden-ksize', type=int, default=7,
                         help='Kernel size of hidden layers convolutions')
 
-    parser.add_argument('--data-channels', type=int, default=3,
-                        help='Number of data channels')
     parser.add_argument('--color-levels', type=int, default=2,
                         help='Number of levels to quantisize value of each channel of each pixel into')