About well formed formula Axiom of specification is schema because it talks about definite condition(or wff) which use notion of finite but this again we define from sets. But in logic we defined wff using consept of tuple and finite like thing. I have just started studying logic. And I am confused about using n-tuple or n-ary relation. Is n-tuple same as we defined ins set theory or defing them separately. If separately then how? 
And how we define definite condition in set theory?
 A: I think I may get some of the confusion behind the OP. In set theory, we generally use first-order logic in order to formulate the axioms (this is more visible with regards to the axiom schema of comprehension and replacement). So, in a sense, set theory "depends" (I'm using the word rather loosely) on a background logic. However, in order to set up our logic, we use notions which are, on the face of it, set-theoretical, e.g., recursive definitions, etc. Therefore, it can look like, to the beginner, that we're caught in a circle: set theory "depends" on logic, yet logic "depends" on set theory.
One way to do away with this impression is simply to be very clear about the meta-theory that we are using (something that most textbooks don't spend too much time on). The meta-theory is simply the background logic that we will use in order to set up our formal system. In this case, first-order logic (FOL, for short) is the meta-theory we're going to use to set up, say, ZFC. This means that every assertion in ZFC will be made using only resources drawn from FOL. In turn, we can formulate all the syntax of FOL using only finitistic notions that, crucially, do not depend on the concept of set (to be more specific, using only primitive recursive arithmetic). So, for instance, instead of defining the set of all formulas by an inductive, set-theoretic definition, we use recursive rules to that allow us to decide when a given sequence of symbols is a formula. Compare, for instance, these two definitions of the sentences of a given propositional language (taken from Chang & Keisler's Model Theory), which illustrate well the difference between the two approaches:
First definition: The set of all sentences of a language $L$ is the least set $\Sigma$ of finite sequences of symbols of $L$ such that each sentence symbol $S$ belongs to $\Sigma$ and, whenever $\psi, \theta$  are in $\Sigma$, then $(\neg \psi), (\psi \wedge \theta)$ belong to $\Sigma$.
Second definition: (i) Every sentence symbol $S$ is a sentence; (ii) If $\phi$ is a sentence, then $(\neg \phi)$ is a sentence; (iii) If $\phi, \psi$ are sentences, then $(\phi \wedge \psi)$ are sentences; (iv) A finite sequence of symbols is a sentence only if it can be shown to be a sentence by a finite number of applications of (i)-(iii).
Now, one could object: the second definition also depends on set theory, as it uses the crucial notion of "finite". While it's true that this notion can be understood in the set-theoretical way (thus smuggling set theory to our definition), one need not understand it this way. Instead, one may take it as a proxy for, say, $n$ times, where $n$ is interpreted in terms of a successive series (i.e. start with "$0$ times" and reiterate the operation one time, then another, then another, etc.). In short, instead of using set-theoretical notions, one uses recursive notions.
Obviously, one may still translate these recursive notions into set-theoretic terms, but that does not mean that one depends on the other. 
A: From Wikipedia:

One instance of the schema is included for each formula $φ$ in the language of set theory with free variables among $x$, $w_1$, $w_2$, ..., $w_n$, $A$. So $B$ is not free in $φ$. In the formal language of set theory, the axiom schema is:
$$\forall w_1,\ldots,w_n \, \forall A \, \exists B \, \forall x \, ( x \in B \Leftrightarrow [ x \in A  \wedge \varphi(x, w_1, \ldots, w_n , A) ] )$$

That's it. The condition in the case of the axiom schema of specification is that you have to have a wff like $φ$. Other notions (such as the notion of free variable, etc.) are parts of the machinery of the first-order logic. ZFC (i.e. the traditional formal system used to capture the truths of informal set theory) is a first-order theory after all, roughtly meaning that it's "built on" FOL.
But at the end of the day, $φ$ really is just a string of symbols.
ADDENDUM: Note that when we talk about variables in math, we are talking about abstract objects - e.g. in the equation x2−1=0 the symbol 'x' denotes an abstract object, i.e. some real number. In logic, though, variables are always just strings of symbols that match some made-up rules. You dictate which strings count as variables. I.e., in ZFC you can choose the symbol 'a' followed by a decimal numeral to be a variable. If you choose to do so, then the string 'a3423924' will be a variable, while the string 'dfskjsk;ghjkdhjkldfshdf' will not.
