Can the OOB estimate of error in randomforest model be realiably used to estimate the accuracy of the model?


While trying randomForest algorithm to predict a categorical variable in a problem, I looked at the model characteristics by typing the model name and I get

Type of random forest: classification
Number of trees: 100
No. of variables tried at each split: 5

    OOB estimate of  error rate: 19.08%

Each time I try with some different set of features I get a different error rate and I saw that lesser the error, more is the accuracy of my model. Can this OOB estimate of error be used as an estimate of the predicting power of a randomforest model always? This would save a lot of submission chances per day in competitions!