Quantum computation and information project, ERATO, Japan Science and Technology Corporation, Daini Hongo White Building 201, Hongo 5-28-3, Bunkyo-ku, Tokyo 113-0033, Japan.Quantum cloning machine for arbitrary mixed states in symmetric subspace is proposed. This quantum cloning machine can be used to copy part of the output state of another quantum cloning machine and is useful in quantum computation and quantum information. The shrinking factor of this quantum cloning achieves the well-known upper bound. When the input is identical pure states, two different fidelities of this cloning machine are optimal.A quantum state cannot be cloned exactly because of the no-cloning theorem [1]. However, quantum cloning approximately (or probablisticaly) is necessary in quantum computation and quantum information [2]. Suppose we have the following task: we have a pure unknown quantum state in 2-level system (qubit) |Ψ . We need one copy of this quantum state to perform one quantum computation. But we do not need it to be exactly the original one. A copy of |Ψ with fidelity of at least 7/9 can give a reliable result. And also we need another 3 identical quantum states each with the fidelity of at least 79/108 to perform another reliable quantum computation. If the fidelities of the quantum states are less than the demanding fidelities, the quantum computation will not be reliable any more. This is certainly a simple and rather practical quantum cloning task. However, we still cannot reach this simple goal by the present available optimal quantum cloning machines.Let's next analyze why the present available quantum cloning machines fail to do this work. 1, First we try to use the 1 to 4 optimal quantum cloning machine proposed by Gisin and Massar [3] to do this work. By this cloning machine, we can copy |Ψ to 4 identical quantum states each with fidelity 3/4 which is larger than 79/108 but less than 7/9. That means we can obtain a reliable result in the second quantum computation but we cannot have a reliable result in the first quantum computation. 2, We may use first the 1 to 2 cloning machine which is proposed by Buzek and Hillery [4,5]. With one output state doing the first quantum computation, then we use another quantum state as input and use the 1 to 3 Gisin-Massar cloning machine to create another 3 identical quantum states. One can find the quantum state of the first cloning machine can achieve the fidelity 5/6 which is better than the demanding fidelity 7/9. However, the 3 identical quantum states can only achieve the fidelity 37/54 which is lower than the demanding fidelity. So, we cannot finish our task by using this method. 3, One is perhaps tempted to use Cerf's asymmetric quantum cloning machine [6] to do this work. The advantage of using Cerf's asymmetric cloning machine is that we can let one quantum state achieve the fidelity 7/9 while another one still has the optimal fidelty since this cloning machine achieve the bound of the no-cloning theorem proposed by Cerf. Then we use the Gisin-Massar 1 to 3 clonin...