diff --git a/scikits/learn/cross_val.py b/scikits/learn/cross_val.py
index f43c4be24256a..ff485b7c6f8bf 100644
--- a/scikits/learn/cross_val.py
+++ b/scikits/learn/cross_val.py
@@ -11,6 +11,7 @@
 
 from .base import is_classifier, clone
 from .utils.extmath import factorial, combinations
+from .utils.fixes import unique
 from .externals.joblib import Parallel, delayed
 
 ##############################################################################
@@ -257,7 +258,7 @@ def __init__(self, y, k):
         assert k>0, ValueError('cannot have k below 1')
         assert k<n, ValueError('cannot have k=%d greater than the number '
                                'of samples %d' % (k, n))
-        _, y_sorted = np.unique1d(y, return_inverse=True)
+        _, y_sorted = unique(y, return_inverse=True)
         assert k <= np.min(np.bincount(y_sorted))
         self.y = y
         self.k = k
@@ -267,7 +268,7 @@ def __iter__(self):
         k = self.k
         n = y.size
 
-        classes = np.unique(y)
+        classes = unique(y)
 
         idx_c = dict()
         j_c = dict()
@@ -345,12 +346,12 @@ def __init__(self, labels):
 
         """
         self.labels = labels
-        self.n_labels = np.unique(labels).size
+        self.n_labels = unique(labels).size
 
     def __iter__(self):
         # We make a copy here to avoid side-effects during iteration
         labels = np.array(self.labels, copy=True)
-        for i in np.unique(labels):
+        for i in unique(labels):
             test_index  = np.zeros(len(labels), dtype=np.bool)
             test_index[labels==i] = True
             train_index = np.logical_not(test_index)
@@ -412,14 +413,14 @@ def __init__(self, labels, p):
 
         """
         self.labels = labels
-        self.unique_labels = np.unique(self.labels)
+        self.unique_labels = unique(self.labels)
         self.n_labels = self.unique_labels.size
         self.p = p
 
     def __iter__(self):
         # We make a copy here to avoid side-effects during iteration
         labels = np.array(self.labels, copy=True)
-        unique_labels = np.unique(labels)
+        unique_labels = unique(labels)
         n_labels = unique_labels.size
         comb = combinations(range(n_labels), self.p)
 
diff --git a/scikits/learn/svm/sparse/base.py b/scikits/learn/svm/sparse/base.py
index 15c9599ef19d3..c6af7be2a746e 100644
--- a/scikits/learn/svm/sparse/base.py
+++ b/scikits/learn/svm/sparse/base.py
@@ -108,13 +108,13 @@ def fit(self, X, Y, class_weight={}):
         self.support_ = sparse.csr_matrix((self._support_data,
                                            self._support_indices,
                                            self._support_indptr),
-                                          shape=(n_SV, X.shape[1])
+                                          (n_SV, X.shape[1])
                                           )
 
         self.dual_coef_ = sparse.csr_matrix((self._dual_coef_data,
                                              dual_coef_indices,
                                              dual_coef_indptr),
-                                            shape=(n_class, n_SV)
+                                            (n_class, n_SV)
                                             )
         return self
 
diff --git a/scikits/learn/utils/_csgraph.py b/scikits/learn/utils/_csgraph.py
index 63630f151416a..535ac4cd473a4 100644
--- a/scikits/learn/utils/_csgraph.py
+++ b/scikits/learn/utils/_csgraph.py
@@ -1,6 +1,9 @@
 """Compressed Sparse graph algorithms"""
 # Backported from scipy 0.9: scipy.sparse.csgraph
 
+# Some compatibility fixes for scipy 0.6
+# Fabian Pedregosa, October 2010
+
 __docformat__ = "restructuredtext en"
 
 __all__ = ['cs_graph_components']
@@ -9,8 +12,8 @@
 
 from sparsetools import cs_graph_components as _cs_graph_components
 
-from scipy.sparse.csr import csr_matrix
-from scipy.sparse.base import isspmatrix
+from scipy.sparse import csr_matrix
+from scipy.sparse import isspmatrix
 
 _msg0 = 'x must be a symmetric square matrix!'
 _msg1 = _msg0 + '(has shape %s)'