diff --git a/pyChemometrics/ChemometricsPLSDA.py b/pyChemometrics/ChemometricsPLSDA.py
index 35fd766..cc82cfe 100644
--- a/pyChemometrics/ChemometricsPLSDA.py
+++ b/pyChemometrics/ChemometricsPLSDA.py
@@ -144,6 +144,7 @@ def fit(self, x, y, **fit_params):
                 y_scaled = self.y_scaler.fit_transform(dummy_mat)
             elif self.n_classes > 2 and isDummy is True:
                 y_scaled = self.y_scaler.fit_transform(y)
+                dummy_mat = y
             else:
                 if y.ndim == 1:
                     y = y.reshape(-1, 1)
@@ -183,7 +184,7 @@ def fit(self, x, y, **fit_params):
                 else:
                     curr_class_idx = np.where(y[curr_class] == 1)
 
-            self.class_means[curr_class, :] = np.mean(self.scores_t[curr_class_idx])
+                self.class_means[curr_class, :] = np.mean(self.scores_t[curr_class_idx], axis=0)
 
             # Needs to come here for the method shortcuts down the line to work...
             self._isfitted = True
@@ -452,8 +453,7 @@ def predict(self, x):
                 class_pred = np.argmin(np.abs(y_pred - np.array([0, 1])), axis=1)
 
             else:
-                # euclidean distance to mean of class for multiclass PLS-DA
-                # probably better to use a Logistic/Multinomial or PLS-LDA anyway...
+                # Euclidean distance to mean of class for multiclass PLS-DA
                 # project X onto T - so then we can get
                 pred_scores = self.transform(x)
                 # prediction rule - find the closest class mean (centroid) for each sample in the score space