diff --git a/Biobase/TrainingData/Filter.hs b/Biobase/TrainingData/Filter.hs
new file mode 100644
--- /dev/null
+++ b/Biobase/TrainingData/Filter.hs
@@ -0,0 +1,40 @@
+{-# LANGUAGE RecordWildCards #-}
+
+-- | Prospective 'TrainingData' elements need to be filtered as there are a
+-- number of entries which do not provide good training.
+
+module Biobase.TrainingData.Filter where
+
+import Data.List
+import qualified Data.Vector.Unboxed as VU
+
+import Biobase.TrainingData
+import Biobase.TrainingData.Manip
+
+
+
+-- | Filter out elements containing not enough base pairs (in relative terms)
+
+fMinRelPairs :: Maybe Double -> TDmanip -> TDmanip
+fMinRelPairs Nothing x    = x
+fMinRelPairs _ l@(Left _) = l
+fMinRelPairs (Just rel) r@(Right td@(TrainingData{..}))
+  | rel > numps * 2 / lenpri = Left td
+  | otherwise = r
+  where
+    lenpri = genericLength primary
+    numps  = genericLength secondary
+
+-- | Error-checking filter.
+
+fErrorCheck :: TDmanip -> TDmanip
+fErrorCheck l@(Left _) = l
+fErrorCheck r@(Right td@(TrainingData{..}))
+  | any (<0) ixs = Left td -- that one is really strange
+  | any (>=l) ixs = Left td -- indices out of bounds
+  | any ((=='&') . (v VU.!)) ixs = Left td -- index pointing to intramolecular symbol
+  | otherwise = r
+  where
+    ixs = concatMap (\((i,j),_) -> [i,j]) secondary
+    l = length primary
+    v = VU.fromList primary
diff --git a/Biobase/TrainingData/Manip.hs b/Biobase/TrainingData/Manip.hs
new file mode 100644
--- /dev/null
+++ b/Biobase/TrainingData/Manip.hs
@@ -0,0 +1,19 @@
+{-# LANGUAGE RecordWildCards #-}
+
+module Biobase.TrainingData.Manip where
+
+import Biobase.Secondary.PseudoKnots
+
+import Biobase.TrainingData
+
+
+
+-- | Left elements are filtered out, Right elements are kept.
+
+type TDmanip = Either TrainingData TrainingData
+
+-- | Remove pseudoknots from 'TrainingData'.
+
+removePK rpk td@TrainingData{..}
+  | not rpk   = td
+  | otherwise = td{secondary = removeByCounting secondary}
diff --git a/BiobaseTrainingData.cabal b/BiobaseTrainingData.cabal
--- a/BiobaseTrainingData.cabal
+++ b/BiobaseTrainingData.cabal
@@ -1,5 +1,5 @@
 name:           BiobaseTrainingData
-version:        0.1.0.0
+version:        0.1.1.0
 author:         Christian Hoener zu Siederdissen
 maintainer:     choener@tbi.univie.ac.at
 homepage:       http://www.tbi.univie.ac.at/~choener/
@@ -24,19 +24,25 @@
                 .
                 Note that several features are designed around /extended/ RNA
                 secondary structures.
+                .
+                Now with some filtering and manipulation options.
 
 library
   build-depends:
     base >3 && <5,
     bytestring,
+    either-unwrap,
     iteratee,
+    vector,
     BiobaseDotP,
     BiobaseFR3D,
     BiobaseXNA >= 0.5.0.1
 
   exposed-modules:
     Biobase.TrainingData
+    Biobase.TrainingData.Filter
     Biobase.TrainingData.Import
+    Biobase.TrainingData.Manip
 
   ghc-options:
     -O2
diff --git a/MkTrainingData.hs b/MkTrainingData.hs
--- a/MkTrainingData.hs
+++ b/MkTrainingData.hs
@@ -1,3 +1,4 @@
+{-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE BangPatterns #-}
 {-# LANGUAGE RecordWildCards #-}
 {-# LANGUAGE DeriveDataTypeable #-}
@@ -6,33 +7,44 @@
 
 import System.Console.CmdArgs
 import Data.List as L
+import Data.Either
+import Data.Either.Unwrap
+import Control.Monad (when)
+import Data.Maybe (isJust, fromJust)
 
-import Biobase.Secondary.PseudoKnots
 import qualified Biobase.FR3D as F
 import qualified Biobase.FR3D.Import as F
 import qualified Biobase.RNAstrand as R
 import qualified Biobase.RNAstrand.Import as R
 
 import Biobase.TrainingData
+import Biobase.TrainingData.Filter
+import Biobase.TrainingData.Manip
 
 
 
 data Options
   -- | FR3D parses all files within a directory and all sub-directories.
   = FR3D
-    { removepk :: Bool
-    , fromdir  :: FilePath
+    { removepk  :: Bool
+    , fromdir   :: FilePath
+    , errorFile :: Maybe FilePath
+    , relativePairs :: Maybe Double
     }
   -- | RNAstrand reads from one file
   | RNAstrand
-    { removepk :: Bool
-    , fromfile :: FilePath
+    { removepk  :: Bool
+    , fromfile  :: FilePath
+    , errorFile :: Maybe FilePath
+    , relativePairs :: Maybe Double
     }
   deriving (Show,Data,Typeable)
 
 fr3d = FR3D
-  { removepk = False &= help "removes pseudoknots from the entry using a heuristic algorithm"
-  , fromdir  = "./"  &= args
+  { removepk  = False &= help "removes pseudoknots from the entry using a heuristic algorithm"
+  , fromdir   = "./"  &= args
+  , errorFile = def   &= help "put TrainingData which falls through the filter in this file (default: disabled)"
+  , relativePairs = def &= help "Keep only TrainingData with that fraction of basepairs."
   }
 
 rnastrand = RNAstrand
@@ -48,7 +60,19 @@
 
 run FR3D{..} = do
   xs <- F.fromDir fromdir
-  mapM_ print $ map (removePK removepk . mkTrainingData . removeBIF . F.linearizeFR3D) xs
+  let (ys :: [TDmanip]) = id
+          . map (fErrorCheck)                 -- basic error-checking
+          . map (fMinRelPairs relativePairs)  -- filtering out trainingdata with too few pairs
+          . map (fmap (removePK removepk))    -- remove pseudoknots from trainingdata
+          . map (fmap (mkTrainingData . removeBIF . F.linearizeFR3D)) -- basic conversions
+          . map Right
+          $ xs
+  let (ls,rs) = partition isLeft ys
+  when (isJust errorFile) $ do
+    writeFile (fromJust errorFile) . unlines . map (show . fromLeft) $ ls
+  mapM_ (print . fromRight) rs
+  return ()
+  -- mapM_ print $ map (removePK removepk . mkTrainingData . removeBIF . F.linearizeFR3D) xs
 
 -- | RNAstrand importer
 
@@ -65,8 +89,3 @@
     | x == "bif" = False
     | otherwise  = True
 
--- | Remove pseudoknots
-
-removePK rpk td@TrainingData{..}
-  | not rpk   = td
-  | otherwise = td{secondary = removeByCounting secondary}