How to drop missing values in pandas

Dropping missing values in pandas removes records or fields that cannot be used safely in a cleanup, analysis, or export step. DataFrame.dropna() is the direct cleanup method when incomplete rows should be excluded instead of filled with replacement values.

dropna() returns a new DataFrame by default, so the original variable remains available for comparison unless the result is assigned back. Row cleanup is the default; column cleanup requires axis=“columns” and should usually target fields that are completely empty or outside the analysis scope.

Choose the rule before deleting data. Use subset for required fields, thresh when a row needs a minimum number of present values, and how=“all” when only fully empty rows or columns should be removed.

1. Start the Python 3 interpreter in an environment with pandas installed.

   $ python3
   Python 3.13.14
   >>>

2. Import pandas and NumPy.

   >>> import pandas as pd
   >>> import numpy as np

3. Create or load the DataFrame that needs missing-value cleanup.

   >>> df = pd.DataFrame(
   ...     {
   ...         "order_id": pd.Series(["1001", "1002", "1003", None, "1005"], dtype="string"),
   ...         "customer": pd.Series(["Ava", "Ben", "Cy", "Dana", "Eli"], dtype="string"),
   ...         "region": pd.Series(["EMEA", "APAC", pd.NA, "EMEA", "AMER"], dtype="string"),
   ...         "total": [125.50, np.nan, 89.00, 42.75, 0.00],
   ...         "shipped_at": [
   ...             pd.Timestamp("2026-06-01"),
   ...             pd.NaT,
   ...             pd.Timestamp("2026-06-03"),
   ...             pd.Timestamp("2026-06-04"),
   ...             pd.NaT,
   ...         ],
   ...         "legacy_note": [np.nan, np.nan, np.nan, np.nan, np.nan],
   ...     }
   ... )

   Use the same variable name for an imported CSV, Excel, SQL, or Parquet DataFrame when cleaning real data.

4. Display the starting rows before dropping anything.

   >>> df
     order_id customer region   total shipped_at  legacy_note
   0     1001      Ava   EMEA  125.50 2026-06-01          NaN
   1     1002      Ben   APAC     NaN        NaT          NaN
   2     1003       Cy   <NA>   89.00 2026-06-03          NaN
   3     <NA>     Dana   EMEA   42.75 2026-06-04          NaN
   4     1005      Eli   AMER    0.00        NaT          NaN

5. Count missing values by column.

   >>> df.isna().sum()
   order_id       1
   customer       0
   region         1
   total          1
   shipped_at     2
   legacy_note    5
   dtype: int64

   dropna() follows pandas missing-value detection. None, NaN, NaT, and pd.NA are missing; empty strings remain ordinary values unless they are converted first.

6. Set the columns that must be present for a row to stay.

   >>> required = ["order_id", "total"]

7. Drop rows that are missing any required column.

   >>> clean_orders = df.dropna(subset=required, ignore_index=True)
   >>> clean_orders
     order_id customer region  total shipped_at  legacy_note
   0     1001      Ava   EMEA  125.5 2026-06-01          NaN
   1     1003       Cy   <NA>   89.0 2026-06-03          NaN
   2     1005      Eli   AMER    0.0        NaT          NaN

   subset=required checks only those columns. ignore_index=True renumbers the remaining rows from zero after the drop.

8. Verify the required columns no longer contain missing cells.

   >>> clean_orders[required].isna().sum()
   order_id    0
   total       0
   dtype: int64

9. Keep rows that have at least four non-missing values when row completeness matters more than a fixed required-column list.

   >>> df.dropna(thresh=4)
     order_id customer region   total shipped_at  legacy_note
   0     1001      Ava   EMEA  125.50 2026-06-01          NaN
   2     1003       Cy   <NA>   89.00 2026-06-03          NaN
   3     <NA>     Dana   EMEA   42.75 2026-06-04          NaN
   4     1005      Eli   AMER    0.00        NaT          NaN

   thresh=4 keeps rows with at least four present cells. Do not combine thresh and how in the same dropna() call.

10. Remove columns that are entirely missing.

    >>> df.dropna(axis="columns", how="all")
      order_id customer region   total shipped_at
    0     1001      Ava   EMEA  125.50 2026-06-01
    1     1002      Ben   APAC     NaN        NaT
    2     1003       Cy   <NA>   89.00 2026-06-03
    3     <NA>     Dana   EMEA   42.75 2026-06-04
    4     1005      Eli   AMER    0.00        NaT

    axis=“columns”, how=“any” removes every column that has even one missing cell. Use it only when partially populated columns should be discarded.