Handling NA values in apply and unique

You were very, very close in your initial solution. But as Aniko remarked, you have to remove NA values before you can use unique.

An example where we first create a similar data.frame and then use apply() as you did -- but with an additional anonymous function that is used to combine na.omit() and unique():

R> DF <- t(data.frame(foo=sample(c(NA, "Foo"), 5, TRUE), 
                      bar=sample(c(NA, "Bar"), 5, TRUE)))
R> DF
    [,1]  [,2] [,3]  [,4]  [,5] 
foo "Foo" NA   "Foo" "Foo" "Foo"
bar NA    NA   NA    "Bar" "Bar"
R> apply(DF, 1, function(x) unique(na.omit(x)))
  foo   bar 
"Foo" "Bar" 

unique does not appear to have an na.rm argument, but you can remove the missing values yourself before calling it:

A <- matrix(c(NA,"A","A",
             "B", NA, NA,
              NA, NA, "C"), nr=3, byrow=TRUE)
apply(A, 1, function(x)unique(x[!is.na(x)]))

gives

[1] "A" "B" "C"