Error Histogram in Artifical Neural Network Skewed - cause of poor success rate?

Hello,

I am running a Artificial Neural Network algorithm in Matlab using the following code that is found here :

    train_target = zeros(40,200);
    for row = 1 : size(train_target, 1)
      col = 5*(row-1)+1;
      train_target(row, col:col+4) = 1;
    end

    %%%%%%%%Creating Network and Training Them %%%%%%%%%%%
    rand('seed', 491218382)
    net = patternnet([90]);
    net.divideFcn = 'divideint';
    net.divideParam.trainRatio =100/100;%%only training
    net.divideParam.valRatio = 00/100;
    net.divideParam.testRatio = 00/100;
    net.trainParam.goal=1e-25;
    for i=2:2%%%set the activation function linear for the output layer
        net.layers{i}.transferFcn = 'purelin';
    end

    % Train network
    [net,tr] = trainrp(net,train_images,train_target);

My success rate is stuck to around 20% at best case. I also found that the Error Histogram is skewed like this:

Can you tell me what might be the issue that is causing the poor success rate?

As a background:

1. I have 40 subjects, with 5 images per subjects (= total of 200 Training images)
2. I used PCA to extract the features from the images.
3. I then ran the code using the test images that was projected as per the PCA algorithm.

Any help is greatly appreciated!