diff --git a/src/NativeProviders/MKL/capabilities.cpp b/src/NativeProviders/MKL/capabilities.cpp
index 21502a82..159d38c3 100644
--- a/src/NativeProviders/MKL/capabilities.cpp
+++ b/src/NativeProviders/MKL/capabilities.cpp
@@ -1,4 +1,5 @@
 #include "wrapper_common.h"
+#include "mkl.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -40,7 +41,8 @@ extern "C" {
 #endif
 
 		// COMMON/SHARED
-		case 64: return 4;	// revision
+		case 64: return 5; // revision
+		case 65: return 1; // numerical consistency, precision and accuracy modes
 
 		// LINEAR ALGEBRA
 		case 128: return 1;	// basic dense linear algebra
@@ -56,6 +58,24 @@ extern "C" {
 		}
 	}
 
+	DLLEXPORT void set_consistency_mode(const MKL_INT mode)
+	{
+		mkl_cbwr_set(mode);
+	}
+
+	DLLEXPORT void set_vml_mode(const MKL_UINT mode)
+	{
+		vmlSetMode(mode);
+	}
+
+	/* Obsolete, will be dropped in the next revision */
+	DLLEXPORT void SetImprovedConsistency(void)
+	{
+		// set improved consistency for MKL and vector functions
+		mkl_cbwr_set(MKL_CBWR_COMPATIBLE);
+		vmlSetMode(VML_HA | VML_DOUBLE_CONSISTENT);
+	}
+
 #ifdef __cplusplus
 }
 #endif /* __cplusplus */
diff --git a/src/NativeProviders/MKL/lapack.cpp b/src/NativeProviders/MKL/lapack.cpp
index 1ae473c2..1ad5a353 100644
--- a/src/NativeProviders/MKL/lapack.cpp
+++ b/src/NativeProviders/MKL/lapack.cpp
@@ -750,11 +750,4 @@ extern "C" {
 			return eigen_complex_factor<MKL_Complex16>(n, a, vectors, values, d, LAPACKE_zgees, LAPACKE_ztrevc);
 		}
     }
-
-    DLLEXPORT void SetImprovedConsistency(void)
-    {
-        // set improved consistency for mkl and vector functions
-        mkl_cbwr_set(MKL_CBWR_COMPATIBLE);
-        vmlSetMode(VML_HA|VML_DOUBLE_CONSISTENT);
-    }
-}
\ No newline at end of file
+}
diff --git a/src/Numerics/Providers/LinearAlgebra/Mkl/MklLinearAlgebraProvider.cs b/src/Numerics/Providers/LinearAlgebra/Mkl/MklLinearAlgebraProvider.cs
index dcff254d..8cc065a4 100644
--- a/src/Numerics/Providers/LinearAlgebra/Mkl/MklLinearAlgebraProvider.cs
+++ b/src/Numerics/Providers/LinearAlgebra/Mkl/MklLinearAlgebraProvider.cs
@@ -4,7 +4,7 @@
 // http://github.com/mathnet/mathnet-numerics
 // http://mathnetnumerics.codeplex.com
 //
-// Copyright (c) 2009-2013 Math.NET
+// Copyright (c) 2009-2014 Math.NET
 //
 // Permission is hereby granted, free of charge, to any person
 // obtaining a copy of this software and associated documentation
@@ -34,6 +34,39 @@ using System;
 
 namespace MathNet.Numerics.Providers.LinearAlgebra.Mkl
 {
+    /// <summary>
+    /// Consistency vs. performance trade-off between runs on different machines.
+    /// </summary>
+    public enum MklConsistency : int
+    {
+        /// <summary>Consistent on the same CPU only (maximum performance)</summary>
+        Auto = 2,
+        /// <summary>Consistent on Intel and compatible CPUs with SSE2 support (maximum compatibility)</summary>
+        Compatible = 3,
+        /// <summary>Consistent on Intel CPUs supporting SSE2 or later</summary>
+        SSE2 = 4,
+        /// <summary>Consistent on Intel CPUs supporting SSE4.2 or later</summary>
+        SSE4_2 = 8,
+        /// <summary>Consistent on Intel CPUs supporting AVX or later</summary>
+        AVX = 9,
+        /// <summary>Consistent on Intel CPUs supporting AVX2 or later</summary>
+        AVX2 = 10
+    }
+
+    [CLSCompliant(false)]
+    public enum MklAccuracy : uint
+    {
+        Low = 0x1,
+        High = 0x2
+    }
+
+    [CLSCompliant(false)]
+    public enum MklPrecision : uint
+    {
+        Single = 0x10,
+        Double = 0x20
+    }
+
     /// <summary>
     /// Intel's Math Kernel Library (MKL) linear algebra provider.
     /// </summary>
@@ -44,17 +77,37 @@ namespace MathNet.Numerics.Providers.LinearAlgebra.Mkl
         bool _nativeX64 = false;
         bool _nativeIA64 = false;
 
-        /// <param name="bitConsistent">If true improves MKL Consistency to get bit consistent results on repeated identical calculations</param>
-        public MklLinearAlgebraProvider(bool bitConsistent = false)
+        readonly MklConsistency _consistency;
+        readonly MklPrecision _precision;
+        readonly MklAccuracy _accuracy;
+
+        /// <param name="consistency">
+        /// Sets the desired bit consistency on repeated identical computations on varying CPU architectures,
+        /// as a trade-off with performance.
+        /// </param>
+        /// <param name="precision">VML optimal precision and rounding.</param>
+        /// <param name="accuracy">VML accuracy mode.</param>
+        [CLSCompliant(false)]
+        public MklLinearAlgebraProvider(
+            MklConsistency consistency = MklConsistency.Auto,
+            MklPrecision precision = MklPrecision.Double,
+            MklAccuracy accuracy = MklAccuracy.High)
         {
-            if (bitConsistent)
-            {
-                SafeNativeMethods.SetImprovedConsistency();
-            }
+            _consistency = consistency;
+            _precision = precision;
+            _accuracy = accuracy;
+        }
+
+        public MklLinearAlgebraProvider()
+        {
+            _consistency = MklConsistency.Auto;
+            _precision = MklPrecision.Double;
+            _accuracy = MklAccuracy.High;
         }
 
         /// <summary>
-        /// Initialize and verify that the provided is indeed available. If not, fall back to alernatives like the managed provider
+        /// Initialize and verify that the provided is indeed available.
+        /// If calling this method fails, consider to fall back to alternatives like the managed provider.
         /// </summary>
         public override void InitializeVerify()
         {
@@ -91,6 +144,12 @@ namespace MathNet.Numerics.Providers.LinearAlgebra.Mkl
             {
                 throw new NotSupportedException("MKL Native Provider too old or not compatible (2).");
             }
+
+            if (SafeNativeMethods.query_capability(65) > 0)
+            {
+                SafeNativeMethods.set_consistency_mode((int)_consistency);
+                SafeNativeMethods.set_vml_mode((uint)_precision | (uint)_accuracy);
+            }
         }
 
         public override string ToString()
diff --git a/src/Numerics/Providers/LinearAlgebra/Mkl/SafeNativeMethods.cs b/src/Numerics/Providers/LinearAlgebra/Mkl/SafeNativeMethods.cs
index b6dfc59d..02f9531f 100644
--- a/src/Numerics/Providers/LinearAlgebra/Mkl/SafeNativeMethods.cs
+++ b/src/Numerics/Providers/LinearAlgebra/Mkl/SafeNativeMethods.cs
@@ -49,6 +49,12 @@ namespace MathNet.Numerics.Providers.LinearAlgebra.Mkl
         [DllImport(DllName, ExactSpelling = true, SetLastError = false, CallingConvention = CallingConvention.Cdecl)]
         internal static extern int query_capability(int capability);
 
+        [DllImport(DllName, ExactSpelling = true, SetLastError = false, CallingConvention = CallingConvention.Cdecl)]
+        internal static extern void set_consistency_mode(int mode);
+
+        [DllImport(DllName, ExactSpelling = true, SetLastError = false, CallingConvention = CallingConvention.Cdecl)]
+        internal static extern void set_vml_mode(uint mode);
+
         #region BLAS
 
         [DllImport(DllName, ExactSpelling = true, SetLastError = false, CallingConvention = CallingConvention.Cdecl)]
@@ -103,9 +109,6 @@ namespace MathNet.Numerics.Providers.LinearAlgebra.Mkl
 
         #region LAPACK
 
-        [DllImport(DllName, ExactSpelling = true, SetLastError = false, CallingConvention = CallingConvention.Cdecl)]
-        internal static extern void SetImprovedConsistency();
-
         [DllImport(DllName, ExactSpelling = true, SetLastError = false, CallingConvention = CallingConvention.Cdecl)]
         internal static extern float s_matrix_norm(byte norm, int rows, int columns, [In] float[] a, [In, Out] float[] work);