Is there a way to iterate over a dictionary?

Yes, NSDictionary supports fast enumeration. With Objective-C 2.0, you can do this:

// To print out all key-value pairs in the NSDictionary myDict
for(id key in myDict)
    NSLog(@"key=%@ value=%@", key, [myDict objectForKey:key]);

The alternate method (which you have to use if you're targeting Mac OS X pre-10.5, but you can still use on 10.5 and iPhone) is to use an NSEnumerator:

NSEnumerator *enumerator = [myDict keyEnumerator];
id key;
// extra parens to suppress warning about using = instead of ==
while((key = [enumerator nextObject]))
    NSLog(@"key=%@ value=%@", key, [myDict objectForKey:key]);

This is iteration using block approach:

    NSDictionary *dict = @{@"key1":@1, @"key2":@2, @"key3":@3};

    [dict enumerateKeysAndObjectsUsingBlock:^(id key, id obj, BOOL *stop) {
        NSLog(@"%@->%@",key,obj);
        // Set stop to YES when you wanted to break the iteration.
    }];

With autocompletion is very fast to set, and you do not have to worry about writing iteration envelope.

The block approach avoids running the lookup algorithm for every key:

[dict enumerateKeysAndObjectsUsingBlock:^(id key, id value, BOOL* stop) {
  NSLog(@"%@ => %@", key, value);
}];

Even though NSDictionary is implemented as a hashtable (which means that the cost of looking up an element is O(1)), lookups still slow down your iteration by a constant factor.

My measurements show that for a dictionary d of numbers ...

NSMutableDictionary* dict = [NSMutableDictionary dictionary];
for (int i = 0; i < 5000000; ++i) {
  NSNumber* value = @(i);
  dict[value.stringValue] = value;
}

... summing up the numbers with the block approach ...

__block int sum = 0;
[dict enumerateKeysAndObjectsUsingBlock:^(NSString* key, NSNumber* value, BOOL* stop) {
  sum += value.intValue;
}];

... rather than the loop approach ...

int sum = 0;
for (NSString* key in dict)
  sum += [dict[key] intValue];

... is about 40% faster.

EDIT: The new SDK (6.1+) appears to optimise loop iteration, so the loop approach is now about 20% faster than the block approach, at least for the simple case above.