Using the object literal is exactly what I would do. A lot of people miss this technique a lot of the time, opting instead for typical array walks as the original code that you showed. The only optimization would be to avoid the arr.length lookup each time. Other than that, O(n) is about as good as you get for uniqueness and is much better than the original O(n^2) example.

function unique(arr) {
    var hash = {}, result = [];
    for ( var i = 0, l = arr.length; i < l; ++i ) {
        if ( !hash.hasOwnProperty(arr[i]) ) { //it works with objects! in FF, at least
            hash[ arr[i] ] = true;
            result.push(arr[i]);
        }
    }
    return result;
}

// * Edited to use hasOwnProperty per comments

Time complexities to summarize

  f()    | unsorted | sorted | objects | scalar | library
____________________________________________________________
unique   |   O(n)   |  O(n)  |   no    |  yes   |    n/a
original |  O(n^2)  | O(n^2) |   yes   |  yes   |    n/a
uniq     |  O(n^2)  |  O(n)  |   yes   |  yes   | Prototype
_.uniq   |  O(n^2)  |  O(n)  |   yes   |  yes   | Underscore

As with most algorithms, there are trade offs. If you are only sorting scalar values, you re modifications to the original algorithm give the most optimal solution. However, if you need to sort non-scalar values, then using or mimicking the uniq method of either of the libraries discussed would be your best choice.

I think your version won t work when you ll have objects or function in the array that give string representation like [Object object]. Because you can only have strings as keys in objects (in the "hash" object here). You ll need to loop into the result array to find if the new entry already exists. It will still be faster than the first method.

Prototype JS has a "uniq" method, you may get inspiration from it.

fun with fun (ctional)

function uniqueNum(arr) {
    return Object.keys(arr.reduce(
        function(o, x) {o[x]=1; return o;}, {})).map(Number);
}  

I m not an algorithm expert by any means, but I ve been keeping an eye on underscore.js. They have this as a function called uniq:

http://documentcloud.github.com/underscore/#uniq

I looked at the code in their library, and copied it here for reference (not my code, this code belongs to underscore.js):

// Produce a duplicate-free version of the array. If the array has already
// been sorted, you have the option of using a faster algorithm.
_.uniq = function(array, isSorted) {
    return _.reduce(array, [], function(memo, el, i) {
        if (0 == i || (isSorted === true ? _.last(memo) != el : !_.include(memo, el))) memo.push(el);
        return memo;
    });
};

EDIT: You need to walk through the rest of the underscore.js code, and I almost took this code out because of it. I left the code snippet in just in case this was still useful.

Unfortunately JS objects have no identity accessible from the language - as other posters have mentioned, using objects as keys in a dictionary will fail when different objects have equal string representations and there is no id() function in the language.

There is a way to avoid the O(n^2) all-pairs check for === identity if you can modify the objects. Pick a random string, walk the array once to check that no object has a property by that name, then just do arr[i][randomPropertyName]=1 for each i. If the next object in the array already has that property, then it is a duplicate.

Unfortunately, the above will only work for modifiable objects. It fails for array values that don t allow property setting (e.g. integers, 42[ random ]=1 just doesn t work :( )